### Install MFA from GitHub

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install MFA directly from its GitHub repository.

```bash
pip install git+https://github.com/MontrealCorpusTools/Montreal-Forced-Aligner.git
```

--------------------------------

### Install MFA from Source

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install MFA from its source repository after cloning. Ensure Kaldi and other dependencies are on the system path.

```bash
python setup.py install
```

```bash
pip install -e .
```

```bash
python setup.py develop
```

--------------------------------

### Install MFA from PyPI

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install the latest stable version of MFA from the Python Package Index.

```bash
pip install montreal-forced-aligner
```

--------------------------------

### Install and Verify MFA

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Install MFA using conda-forge and verify the installation. Optionally, install SpeechBrain support and set up Docker.

```bash
conda create -n aligner -c conda-forge montreal-forced-aligner
conda activate aligner

mfa version

mfa_update

conda install pytorch torchaudio cpuonly -c pytorch
pip install speechbrain

docker image pull mmcauliffe/montreal-forced-aligner:latest
docker run -it -v /path/to/data:/data mmcauliffe/montreal-forced-aligner:latest
# Inside container:
mfa model download acoustic english_mfa
mfa align /data/corpus /data/english_us_mfa.dict english_mfa /data/output
```

--------------------------------

### start_server

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/reference/server/index.rst

Starts the MFA server.

```APIDOC
## start_server

### Description
Starts the MFA server.

### Method
Not specified (likely a function call within the SDK)

### Endpoint
Not applicable (function call)

### Parameters
No parameters are explicitly documented.
```

--------------------------------

### Dockerfile for Installing MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

A sample Dockerfile demonstrating how to install MFA within a custom container using mambaforge. It sets up a user and environment for MFA.

```docker
FROM condaforge/mambaforge:22.11.1-4 as build

RUN mkdir -p /mfa
RUN mamba create -p /env -c conda-forge montreal-forced-aligner

RUN useradd -ms /bin/bash mfauser
RUN chown -R mfauser /mfa
RUN chown -R mfauser /env
USER mfauser
ENV MFA_ROOT_DIR=/mfa
RUN conda run -p /env mfa server init

RUN echo "source activate /env && mfa server start" > ~/.bashrc
ENV PATH /env/bin:$PATH
```

--------------------------------

### Install PyTorch (CPU)

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install PyTorch, torchvision, and torchaudio for CPU-only usage. The '-c pytorch' flag specifies the pytorch channel.

```bash
conda install pytorch torchvision torchaudio cpuonly -c pytorch
```

--------------------------------

### Install PyTorch (GPU)

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install PyTorch, torchvision, and torchaudio with CUDA support for GPU usage. '-c pytorch' and '-c nvidia' specify the respective channels.

```bash
conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
```

--------------------------------

### Install Mandarin Dependencies

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Install Mandarin-specific dependencies using pip.

```bash
pip install spacy-pkuseg dragonmapper hanziconv
```

--------------------------------

### Install Japanese Dependencies

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Install Japanese-specific dependencies using conda.

```bash
conda install -c conda-forge spacy sudachipy sudachidict-core
```

--------------------------------

### Verify MFA Installation

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Check if MFA is installed and accessible from the console.

```bash
mfa -h
```

--------------------------------

### Using Per-Speaker Dictionaries in MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/dictionary.rst

Command-line example demonstrating how to use a per-speaker dictionary configuration file with the 'mfa align' command.

```bash
mfa align /path/to/corpus /path/to/speaker_dictionaries.yaml /path/to/acoustic_model.zip /path/to/output
```

--------------------------------

### User and Environment Setup in Dockerfile

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

This snippet from a Dockerfile shows the creation of a dedicated user ('mfauser'), ownership changes for directories, setting the MFA root directory, and initializing the MFA server within the environment.

```docker
RUN useradd -ms /bin/bash mfauser
RUN chown -R mfauser /mfa
RUN chown -R mfauser /env
USER mfauser
ENV MFA_ROOT_DIR=/mfa
RUN conda run -p /env mfa server init
```

--------------------------------

### Phone Set Mapping Example

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/index.rst

An example of a YAML file used to map phones from one set to another. This is required for the 'mfa remap_dictionary' command.

```yaml
aj: AY1
aw: AW1
b: B
...
m̩: AH0 M
...
```

--------------------------------

### Install Mamba

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install mamba, a faster drop-in replacement for conda, in your base conda environment.

```bash
conda install -c conda-forge mamba
```

--------------------------------

### Install SpeechBrain via Pip

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install the SpeechBrain library using pip after ensuring you are in the correct conda environment and have PyTorch installed.

```bash
pip install speechbrain
```

--------------------------------

### Install Anchor Subpackage

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/corpus_creation/anchor.rst

Install the anchor subpackage for MFA using conda. This command installs MFA if it is not already present, along with all required packages for Anchor.

```bash
conda install montreal-forced-aligner[anchor]
```

--------------------------------

### Install MFA in Develop Mode

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/README.md

Install MFA in editable (develop) mode using pip. This allows you to make changes to the source code and have them reflected immediately. The `[dev]` extra installs development dependencies.

```bash
pip install -e .[dev]
```

--------------------------------

### Install Stable 2.1 Release of MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install a specific older version (2.1.7) of MFA along with its compatible versions of openfst and kaldi.

```bash
conda install -c conda-forge montreal-forced-aligner=2.1.7 openfst=1.8.2 kaldi=5.5.1068
```

--------------------------------

### Install Stable 2.0 Release of MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install a specific older version (2.0.6) of MFA along with its compatible versions of openfst and kaldi.

```bash
conda install -c conda-forge montreal-forced-aligner=2.0.6 openfst=1.8.2 kaldi=5.5.1068
```

--------------------------------

### YAML Phone Mapping Format Example

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Example of the YAML format for phone set mapping files. Keys are source phone set symbols, and values are target phone set symbols.

```yaml
# ipa_to_arpa_mapping.yaml:
# p: P
# b: B
# t: T
# d: D
# aɪ: AY
```

--------------------------------

### Manage MFA Database Servers via CLI

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/server/index.rst

Use the 'mfa server' subcommands to manage PostgreSQL servers, including initialization, starting, stopping, and deletion.

```bash
mfa server
```

--------------------------------

### List Acoustic Models

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/models/index.rst

Lists currently saved acoustic models. Initially, this will be empty after installation.

```bash
mfa model list acoustic
```

--------------------------------

### Install Stable 2.2 Release of MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install a specific older version (2.2.17) of MFA along with its compatible versions of openfst and kaldi.

```bash
conda install -c conda-forge montreal-forced-aligner=2.2.17 openfst=1.8.2 kaldi=5.5.1068
```

--------------------------------

### Install Specific MFA Version

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Install a specific version of the Montreal Forced Aligner by appending '=X.X.X' to the package name.

```bash
conda install montreal-forced-aligner=X.X.X
```

--------------------------------

### Install MFA with Conda

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/getting_started.rst

Installs the Montreal Forced Aligner using conda. This involves adding the conda-forge channel, creating a new environment, activating it, and verifying the installation with the help command.

```bash
conda config --add channels conda-forge
conda create -n aligner montreal-forced-aligner
conda activate aligner
mfa --help
```

--------------------------------

### Run MFA Docker Container

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Start an interactive Docker container for MFA, mounting a local directory to '/data' inside the container for data access. The path before the colon is local, and after is inside the container.

```bash
docker run -it -v /path/to/data/directory:/data mmcauliffe/montreal-forced-aligner:latest
```

--------------------------------

### Adapt using reference (gold) alignments to guide adaptation

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

This method uses reference (gold) alignments to guide the model adaptation process.

```APIDOC
## Adapt using reference (gold) alignments to guide adaptation

### Description
Guides the model adaptation process using provided reference (gold) alignments.

### Command Line Usage
```bash
mfa adapt /corpus/target /path/to/dict.dict english_mfa /output/adapted.zip \
    --reference_directory /path/to/gold_alignments
```
```

--------------------------------

### Default Ivector Training Configuration

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/configuration/ivector.rst

This YAML configuration file demonstrates the available options for both DUBM and i-vector training, reflecting the settings used in the MFA 2.0 training regime. It serves as an example for understanding the configuration structure and parameters.

```yaml
features:
  type: "mfcc"
  use_energy: true
  frame_shift: 10

training:
  - dubm:
      num_iterations: 4
      num_gselect: 30
      num_gaussians: 256
      num_iterations_init: 20
  - ivector:
      ivector_dimension: 128
      num_iterations: 10
      gaussian_min_count: 100
      silence_weight: 0.0
      posterior_scale: 0.1
      max_count: 100

```

--------------------------------

### Create Conda Environment for MFA (Windows)

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Create a new conda environment named 'aligner' with Kaldi installed for Windows.

```bash
conda create -n aligner kaldi
```

--------------------------------

### Download Pretrained Models in Docker

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Inside the MFA Docker container, use this command to download pretrained acoustic models, for example, the English model.

```bash
mfa model download acoustic english_mfa
```

--------------------------------

### Install MFA Dependencies via Conda

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/README.md

Alternatively, install MFA dependencies individually using conda. This command lists common dependencies for MFA.

```bash
conda install -c conda-forge python=3.11 kaldi librosa praatio tqdm requests colorama pyyaml pynini openfst baumwelch ngram
```

--------------------------------

### Create Conda Environment for MFA (Linux/macOS)

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Create a new conda environment named 'aligner' with Kaldi and Pynini installed for Linux or macOS.

```bash
conda create -n aligner kaldi pynini
```

--------------------------------

### Update MFA Model CLI

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/changelog/index.md

Example of how to update an MFA model with new pronunciations from a G2Pped file.

```bash
mfa model update /path/to/g2pped_file.txt
```

--------------------------------

### Create Conda Environment and Install MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Use this command to create a new conda environment named 'aligner' and install the Montreal Forced Aligner. The '-c conda-forge' flag specifies the conda-forge channel.

```bash
conda create -n aligner -c conda-forge montreal-forced-aligner
```

--------------------------------

### Install Stable 3.0 Release of MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Update to the latest version within the 3.0 release series of the Montreal Forced Aligner.

```bash
conda update -c conda-forge montreal-forced-aligner
```

--------------------------------

### Adapt a model and save the adapted version

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

This section details how to adapt an existing acoustic model to a target domain and save the adapted model. It includes command-line usage and a Python API example.

```APIDOC
## Adapt a model and save the adapted version

### Description
Adapts an existing acoustic model to a target domain and saves the adapted version.

### Command Line Usage
```bash
mfa adapt /corpus/target_domain \
          /path/to/dict.dict \
          english_mfa \
          /output/adapted_model.zip
```

### Python API
```python
from montreal_forced_aligner.alignment import AdaptingAligner

adapter = AdaptingAligner(
    corpus_directory="/corpus/target",
    dictionary_path="/path/to/dict.dict",
    acoustic_model_path="english_mfa",
)
try:
    adapter.adapt()
    adapter.export_model("/output/adapted_model.zip")
finally:
    adapter.cleanup()
```
```

--------------------------------

### PostgreSQL Configuration for MFA

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/server/index.rst

These are the default PostgreSQL configuration values overridden by MFA during initialization. Tune these further based on your setup and corpus.

```ini
log_min_duration_statement = 5000
enable_partitionwise_join = on
enable_partitionwise_aggregate = on
unix_socket_directories = '/path/to/current/profile/socket_directory'
listen_addresses = ''

maintenance_work_mem = 1GB
work_mem = 128MB
shared_buffers = 256MB
max_connections = 1000
```

--------------------------------

### Example Phone Groupings in YAML

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/implementations/phone_groups.md

This YAML snippet demonstrates how to group phonetically similar phones together. Use this to specify custom phone groups for acoustic modeling.

```yaml
- 
  - t
  - tʷ
  - tʰ
  - tʲ
- 
  - d
  - dʲ
- 
  - ɾ
  - ɾʲ
- 
  - ʔ
```

--------------------------------

### Python API for model adaptation

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

This Python code demonstrates how to adapt an acoustic model using the AdaptingAligner class. It includes setup, adaptation, model export, and cleanup steps. Ensure all paths and model names are correct.

```python
from montreal_forced_aligner.alignment import AdaptingAligner

adapter = AdaptingAligner(
    corpus_directory="/corpus/target",
    dictionary_path="/path/to/dict.dict",
    acoustic_model_path="english_mfa",
)
try:
    adapter.adapt()
    adapter.export_model("/output/adapted_model.zip")
finally:
    adapter.cleanup()
```

--------------------------------

### Default Diarization Configuration

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/configuration/diarization.rst

This is an example of a default configuration file for diarization. It specifies parameters like the clustering algorithm, energy scaling, and segment length.

```yaml
cluster_type: optics
energy_mean_scale: 0.5
max_segment_length: 30
min_pause_duration: 0.05

```

--------------------------------

### Phone mapping YAML for remapping alignments

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/workflows/remap.rst

Example of a phone mapping YAML file for remapping alignment phone tiers. This format is stricter than for dictionary remapping, requiring a single target phone per source phone.

```yaml
aj: AY1
aw: AW1
b: B
bʲ: B
c: K
cʰ: K
cʷ: K
d: D
dʒ: JH
dʲ: D
ej: EY1
f: F
fʲ: F
h: HH
i: IY0
iː: IY1
...

```

--------------------------------

### initialize_server

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/reference/server/index.rst

Initializes a new MFA server.

```APIDOC
## initialize_server

### Description
Initializes a new MFA server.

### Method
Not specified (likely a function call within the SDK)

### Endpoint
Not applicable (function call)

### Parameters
No parameters are explicitly documented.
```

--------------------------------

### Download Dictionary

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/alignment_example.md

Download a pretrained dictionary for a specific language. Ensure the dictionary name is correct for your language.

```bash
mfa model download dictionary english_us_mfa
```

```bash
mfa model download dictionary japanese_mfa
```

```bash
mfa model download dictionary mandarin_china_mfa
```

--------------------------------

### Download Japanese Dictionary

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Use this command to download the pretrained Japanese dictionary for adaptation.

```bash
mfa model download dictionary japanese_mfa
```

--------------------------------

### Download Pretrained Models

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/index.rst

Use these commands to download the necessary acoustic and dictionary models for a specific language. Ensure the 'mfa' command is accessible in your PATH.

```bash
mfa model download acoustic english_us_arpa
mfa model download dictionary english_us_arpa
```

--------------------------------

### Download G2P model for Japanese

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/alignment_example.md

Use this command to download the pretrained G2P model for Japanese.

```bash
mfa model download g2p japanese_mfa
```

--------------------------------

### Install MFA via Conda

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/README.md

Use this command to install the Montreal Forced Aligner using conda-forge. Ensure you are in your desired environment.

```bash
conda install -c conda-forge montreal-forced-aligner
```

--------------------------------

### Download English Dictionary

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Use this command to download the pretrained US English dictionary for adaptation.

```bash
mfa model download dictionary english_us_mfa
```

--------------------------------

### Download G2P model for English

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/alignment_example.md

Use this command to download the pretrained G2P model for US English.

```bash
mfa model download g2p english_us_mfa
```

--------------------------------

### Create Development Environment

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/README.md

Create a development environment for MFA using the provided environment.yml file. This is useful for local development and testing.

```bash
conda env create -n mfa-dev -f environment.yml
```

--------------------------------

### Download Pretrained Acoustic Model and Dictionary

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/index.rst

Download necessary pretrained acoustic models and pronunciation dictionaries for tasks like remapping. Ensure the 'mfa' command is accessible.

```bash
mfa model download acoustic english_us_arpa
mfa model download dictionary english_us_mfa
```

--------------------------------

### Download Japanese JVS Demo Corpus

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Download the Japanese JVS demo corpus for adaptation.

```bash
Download the [Japanese JVS demo corpus](https://github.com/MontrealCorpusTools/japanese-jvs-demo/archive/refs/tags/v1.0.0.tar.gz) and extract it to somewhere on your computer
```

--------------------------------

### Manage MFA Models

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Download, list, inspect, and save pretrained MFA models (acoustic, G2P, dictionaries, language models).

```bash
# List all downloadable acoustic models
mfa model download acoustic

# Download a specific model
mfa model download acoustic english_mfa
mfa model download dictionary english_us_mfa
mfa model download g2p english_us_mfa
mfa model download language_model english_mfa

# Download a specific version of a model
mfa model download acoustic english_mfa --version 3.0.0

# List locally saved models of a given type
mfa model list acoustic
mfa model list dictionary

# Inspect metadata of a downloaded model
mfa model inspect acoustic english_mfa
# Output includes: version, phone_set, features, training data info, etc.

# Save a custom-trained model under a friendly name for reuse
mfa model save acoustic /path/to/my_model.zip --name my_english_model
# Now usable as: mfa align ... my_english_model ...

# Add pronunciations from a supplementary dictionary into a base dictionary
mfa model add_words /path/to/base.dict /path/to/extra_words.dict
```

--------------------------------

### Download English LibriSpeech Demo Corpus

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Download the English LibriSpeech demo corpus for adaptation.

```bash
Download the [English LibriSpeech demo corpus](https://github.com/MontrealCorpusTools/librispeech-demo/archive/refs/tags/v1.0.0.tar.gz) and extract it to somewhere on your computer
```

--------------------------------

### Acoustic-model-guided segmentation

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Segments audio files using an acoustic model to guide the process.

```APIDOC
## segment /corpus/long_recordings /path/to/dict.dict english_mfa /output/segments

### Description
Segments audio files using an acoustic model.

### Method
CLI Command

### Endpoint
mfa segment

### Parameters
#### Path Parameters
- **/corpus/long_recordings** (string) - Required - Directory containing audio files to segment.
- **/path/to/dict.dict** (string) - Required - Path to the dictionary file.
- **english_mfa** (string) - Required - Path to the acoustic model.
- **/output/segments** (string) - Required - Directory for output segment files.
```

--------------------------------

### Download Pretrained Model by Type

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/models/index.rst

Downloads pretrained models for alignment or pronunciation dictionaries. Replace <model_type> with 'acoustic', 'g2p', or 'dictionary'.

```bash
mfa model download <model_type>
```

--------------------------------

### Run MFA Update Utility

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Use the 'mfa_update' utility to fetch the latest versions of MFA, Kalpy, and Kaldi. This is available in MFA version 3.0.5 and later.

```bash
mfa_update
```

--------------------------------

### SpeechBrain-based Transcription

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Transcribes audio files using the SpeechBrain toolkit. Requires speechbrain to be installed.

```bash
mfa transcribe_speechbrain /corpus/audio english /output/transcriptions \
    --architecture wav2vec2
```

--------------------------------

### Download Mandarin Acoustic Model

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Use this command to download the pretrained Mandarin acoustic model for adaptation.

```bash
mfa model download acoustic mandarin_mfa
```

--------------------------------

### SpeechBrain-based transcription

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Transcribes audio files using a SpeechBrain-based model. Requires the speechbrain library to be installed.

```APIDOC
## transcribe_speechbrain /corpus/audio english /output/transcriptions

### Description
Transcribes audio files using a SpeechBrain-based model.

### Method
CLI Command

### Endpoint
mfa transcribe_speechbrain

### Parameters
#### Path Parameters
- **/corpus/audio** (string) - Required - Directory containing audio files.
- **english** (string) - Required - Language of the audio.
- **/output/transcriptions** (string) - Required - Directory for output transcriptions.

#### Query Parameters
- **--architecture** (string) - Optional - Specifies the architecture of the SpeechBrain model (e.g., wav2vec2).
```

--------------------------------

### Download Acoustic Model

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/alignment_example.md

Download a pretrained acoustic model for a specific language. Ensure the model name is correct for your language.

```bash
mfa model download acoustic english_mfa
```

```bash
mfa model download acoustic japanese_mfa
```

```bash
mfa model download acoustic mandarin_mfa
```

--------------------------------

### Uninstall Montreal Forced Aligner (pip)

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Remove any existing pip installation of MFA before upgrading via Conda.

```bash
pip uninstall montreal-forced-aligner
```

--------------------------------

### Download G2P model for Mandarin

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/alignment_example.md

Use this command to download the pretrained G2P model for Mandarin Chinese.

```bash
mfa model download g2p mandarin_china_mfa
```

--------------------------------

### Example French clitic normalization

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/dictionary.rst

Demonstrates how French clitics marked with apostrophes are handled during text normalization and dictionary lookup.

```none
c'est S E
c     S E
c'    S
etait E T E
un    A N
```

```none
c'est un c
c'etait un c
```

```none
S E A N S E
S E T E A N S E
```

--------------------------------

### Download Mandarin China Dictionary

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Use this command to download the pretrained China Mandarin dictionary for adaptation.

```bash
mfa model download dictionary mandarin_china_mfa
```

--------------------------------

### Download Pretrained G2P Model

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/index.rst

Download a pretrained G2P (Grapheme-to-Phoneme) model for generating pronunciation dictionaries. This is a prerequisite for using the 'mfa g2p' command.

```bash
mfa model download g2p english_us_arpa
```

--------------------------------

### Download Mandarin THCHS-30 Demo Corpus

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Download the Mandarin THCHS-30 demo corpus for adaptation.

```bash
Download the [Mandarin THCHS-30 demo corpus](https://github.com/MontrealCorpusTools/mandarin-thchs-30-demo/archive/refs/tags/v1.0.0.tar.gz) and extract it to somewhere on your computer
```

--------------------------------

### Transcriber Class

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/reference/transcription/main.rst

The Transcriber class is the main entry point for using the transcription functionality. It handles the setup and execution of transcription jobs.

```APIDOC
## Transcriber

### Description

The Transcriber class is the primary interface for performing transcription tasks within the Montreal Forced Aligner. It encapsulates the logic for setting up and running transcription processes.

### Usage

Instantiate the Transcriber class and use its methods to manage transcription jobs.

```python
from montreal_forced_aligner.transcription import Transcriber

# Example usage (details of methods and parameters would be further documented)
transcriber = Transcriber(config)
transcriber.transcribe(corpus_path, output_directory)
```

### Methods

(Specific methods like `transcribe`, `train`, etc., would be detailed here with their parameters and return values if documented in the source.)

### Parameters

(Parameters for the `Transcriber` constructor and its methods would be detailed here if documented in the source.)

### Returns

(Return values for `Transcriber` methods would be detailed here if documented in the source.)
```

--------------------------------

### MFA Dictionary Pronunciations

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/dictionary.rst

Example pronunciations generated by MFA for specific words and cutoff patterns, including silence ('spn') and phonetic transcriptions.

```text
<cutoff>   spn
<cutoff-cut>   spn
<cutoff-cut>   kʰ ɐ t
<cutoff-cut>   kʰ ɐ
<cutoff-cut>   kʰ
<cutoff-off>   spn
<cutoff-off>   ɒ f
<cutoff-off>   ɒ
<cutoff-off>   ɑ f
<cutoff-off>   ɑ
```

--------------------------------

### Set Beam Width via YAML Configuration File

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/configuration/index.rst

For more complex configurations, define parameters in a YAML file and reference it using the `--config_path` argument. This allows for specifying multiple settings at once.

```yaml
beam: 1000
```

--------------------------------

### Get Log Likelihood

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/notebooks/debugging_alignments.ipynb

Retrieve the log likelihood for a specific frame and state using the Decodable object. This helps in understanding acoustic scoring.

```python
like = decodable.LogLikelihood(1, 2)
print(like)
```

--------------------------------

### Load and Prepare Alignment Data

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/notebooks/debugging_alignments.ipynb

Load reference phones, features, and training graphs using specified paths. Initialize the GMM aligner with the acoustic model.

```python
reference_phone_archive = RandomAccessInt32VectorVectorReader(generate_read_specifier(reference_phones_path))
reference_phones = [x[0] for x in reference_phone_archive.Value(utterance_id)]
reference_phones[0] = -1

feature_archive = FeatureArchive(
            feat_path,
            lda_mat_file_name=lda_mat_path,
        )
training_graph_archive = FstArchive(fst_path)

training_graph = training_graph_archive[utterance_id]
features = feature_archive[utterance_id]

aligner = GmmAligner(model_path)

print(len(reference_phones))
print(features.NumRows())
```

--------------------------------

### SpeechBrain VAD Segmentation with CUDA

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Segments audio using SpeechBrain's VAD model, leveraging CUDA for GPU acceleration. Requires speechbrain installation.

```bash
mfa segment_vad /corpus/long_recordings /output/segments --speechbrain --cuda
```

--------------------------------

### Align with Reference Phones

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/notebooks/debugging_alignments.ipynb

Perform alignment using reference phones as a guide. This function returns detailed alignment information, including likelihoods and success status.

```python
( 
                alignment,
                words,
                likelihood,
                per_frame_log_likelihoods,
                successful,
                retried,
            ) = gmm_align_reference_phones(
                aligner.transition_model,
                aligner.acoustic_model,
                training_graph,
                features,
                reference_phones,
                acoustic_scale=aligner.acoustic_scale,
                transition_scale=aligner.transition_scale,
                self_loop_scale=aligner.self_loop_scale,
                beam=aligner.beam,
                retry_beam=aligner.retry_beam,
                careful=aligner.careful,
            )
print(alignment[:10])
alignment_phones = [aligner.transition_model.TransitionIdToPhone(x) for x in alignment]
```

--------------------------------

### Define Cot-Caught Merger Rules in YAML

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/implementations/phonological_rules.md

Example of defining phonological rules for the cot-caught merger in English, specifying variants for long/short vowels and different following contexts.

```yaml
rules:
  - following_context: $  # caught-cot merger
    preceding_context: ''
    replacement: ɑː
    segment: ɒː
  - following_context: '[^ɹ]'  # caught-cot merger
    preceding_context: ''
    replacement: ɑː
    segment: ɒː
  - following_context: $  # caught-cot merger
    preceding_context: ''
    replacement: ɑ
    segment: ɒ
  - following_context: '[^ɹ]'  # caught-cot merger
    preceding_context: ''
    replacement: ɑ
    segment: ɒ
```

--------------------------------

### Acoustic-model-guided Segmentation

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Segments audio files using an acoustic model for guidance. Requires a dictionary and acoustic model path.

```bash
mfa segment /corpus/long_recordings /path/to/dict.dict english_mfa \
    /output/segments
```

--------------------------------

### Create Conda Environment with Mamba

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/installation.rst

Use mamba to create a new conda environment named 'aligner' and install the Montreal Forced Aligner. This is generally faster than using conda.

```bash
mamba create -n aligner -c conda-forge montreal-forced-aligner
```

--------------------------------

### MFA Alignment Evaluation Output

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

This is an example of the output from the MFA alignment evaluation process when reference directories are provided. It shows average overlap scores and phone error rates.

```text
INFO     Evaluating alignments...
INFO     Exporting evaluation...
INFO     Average overlap score: 0.010834011956534382
INFO     Average phone error rate: 0.02820097244732577
```

--------------------------------

### Adapt a model and save the adapted version

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Use this command to adapt an existing acoustic model to a target domain and save the adapted model. Ensure the corpus directory, dictionary, acoustic model, and output path are correctly specified.

```bash
mfa adapt /corpus/target_domain \
          /path/to/dict.dict \
          english_mfa \
          /output/adapted_model.zip
```

```bash
mfa adapt /corpus/target /path/to/dict.dict english_mfa /output/adapted.zip \
    --output_directory /output/alignments
```

```bash
mfa adapt /corpus/target /path/to/dict.dict english_mfa /output/adapted.zip \
    --reference_directory /path/to/gold_alignments
```

--------------------------------

### Complex Phonological Rule Examples in YAML

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/implementations/phonological_rules.md

Demonstrates various complex phonological rules including deletion, syllabic 'l' insertion, schwa deletion, consonant cluster simplification, and metathesis.

```yaml
rules:
  - following_context: ''  # deleting d after n
    preceding_context: 'n'
    replacement: ''
    segment: d
  - following_context: '[^ʊɔɝaɛeoæɐɪəɚɑʉɒi].*'  # syllabic l
    preceding_context: ''
    replacement: ɫ̩
    segment: ə ɫ
  - following_context: ''  # schwa deletion
    preceding_context: ''
    replacement: ɹ ə
    segment: 'ə ɹ ə'
  - following_context: ''
    preceding_context: ''
    replacement: dʒ
    segment: d[ʲʷ]? ɹ
  - following_context: $  # ask metathesis
    preceding_context: ''
    replacement: k s
    segment: s k
```

--------------------------------

### configure_pg

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/reference/server/index.rst

Configures the PostgreSQL database for the server.

```APIDOC
## configure_pg

### Description
Configures the PostgreSQL database for the server.

### Method
Not specified (likely a function call within the SDK)

### Endpoint
Not applicable (function call)

### Parameters
No parameters are explicitly documented.
```

--------------------------------

### Complex Configuration with YAML

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/configuration/index.rst

Define multiple advanced settings, including training parameters, punctuation, and beam widths, within a single YAML configuration file for use with MFA.

```yaml
beam: 100
retry_beam: 400

punctuation: ":,."

training:
  - monophone:
      num_iterations: 20
      max_gaussians: 500
      subset: 1000
      boost_silence: 1.25

  - triphone:
      num_iterations: 35
      num_leaves: 2000
      max_gaussians: 10000
      cluster_threshold: -1
      subset: 5000
      boost_silence: 1.25
      power: 0.25
```

--------------------------------

### Remap Japanese to English Dictionary

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/remapping_example.md

Use this command to create a Japanese dictionary with English MFA pronunciations. Download the mapping file first.

```bash
mfa remap dictionary japanese_mfa english_mfa ~/mfa_data/japanese_to_english_phone_mapping.yaml ~/mfa_data/japanese_english.dict
```

--------------------------------

### Python API for language model training

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

This Python code uses `MfaLmCorpusTrainer` to train a language model from a corpus directory. It handles the setup, training, and export of the model. Ensure the corpus directory path is correct.

```python
from montreal_forced_aligner.language_modeling.trainer import MfaLmCorpusTrainer

trainer = MfaLmCorpusTrainer(corpus_directory="/corpus/text_data")
try:
    trainer.setup()
    trainer.train()
    trainer.export_model("/output/my_lm.zip")
finally:
    trainer.cleanup()
```

--------------------------------

### Train a G2P model from a pronunciation dictionary

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Use `mfa train_g2p` to train a new grapheme-to-phoneme model using Pynini or Phonetisaurus. This requires an existing pronunciation dictionary. Options include evaluation on a held-out set and using custom configuration files.

```bash
mfa train_g2p /path/to/english_us_mfa.dict /output/my_g2p_model.zip
```

```bash
mfa train_g2p /path/to/dict.dict /output/my_g2p_model.zip --evaluate
```

```bash
mfa train_g2p /path/to/dict.dict /output/my_g2p_model.zip --phonetisaurus
```

```bash
mfa train_g2p /path/to/dict.dict /output/my_g2p_model.zip \
    --config_path /path/to/g2p_config.yaml
```

--------------------------------

### Import Kalpy Debugging Utilities

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/notebooks/debugging_alignments.ipynb

Import necessary classes and functions from Kalpy for GMM alignment and data handling. Ensure these are available in your environment.

```python
from kalpy.decoder.data import FstArchive
from kalpy.gmm.align import GmmAligner
from kalpy.feat.data import FeatureArchive
from kalpy.utils import generate_read_specifier
from _kalpy.util import RandomAccessInt32VectorVectorReader
from _kalpy.gmm import gmm_align_compiled, gmm_align_reference_phones
from _kalpy.gmm import DecodableAmDiagGmmScaledMasked
```

--------------------------------

### Default Training Configuration (MFA 2.0)

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/configuration/acoustic_modeling.rst

This YAML configuration represents the default training regime for MFA version 2.0, illustrating available configuration options and their progression.

```yaml
beam: 10
retry_beam: 40

features:
  type: "mfcc"
  use_energy: false
  use_pitch: true
  frame_shift: 10

training:
  - monophone:
      subset: 10000
      num_iterations: 40
      max_gaussians: 1000
      boost_silence: 1.25

  - triphone:
      subset: 20000
      num_iterations: 35
      num_leaves: 2000
      max_gaussians: 10000
      cluster_threshold: -1
      boost_silence: 1.25
      power: 0.25

  - lda:
      subset: 20000
      num_leaves: 2500
      max_gaussians: 15000
      num_iterations: 35

  - sat:
      subset: 20000
      num_leaves: 2500
      max_gaussians: 15000
      power: 0.2
      silence_weight: 0.0
      fmllr_update_type: "full"

  - sat:
      subset: 50000
      num_leaves: 4200
      max_gaussians: 40000
      power: 0.2
      silence_weight: 0.0
      fmllr_update_type: "full"

  - pronunciation_probabilities:
      subset: 50000
      silence_probabilities: true

  - sat:
      subset: 150000
      num_leaves: 5000
      max_gaussians: 100000
      power: 0.2
      silence_weight: 0.0
      fmllr_update_type: "full"

  - pronunciation_probabilities:
      subset: 150000
      silence_probabilities: true
      optional: true # Skipped if the corpus is smaller than the subset

  - sat:
      subset: 0
      quick: true # Performs fewer fMLLR estimation
      num_iterations: 20
      num_leaves: 7000
      max_gaussians: 150000
      power: 0.2
      silence_weight: 0.0
      fmllr_update_type: "full"
      optional: true # Skipped if the corpus is smaller than the previous subset

```

--------------------------------

### Python API for G2P word list generation

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

This Python code uses `PyniniWordListGenerator` to create pronunciations from a word list. It allows specifying the number of pronunciations and handles setup, generation, and cleanup. Ensure the G2P model path is correct.

```python
from montreal_forced_aligner.g2p.generator import PyniniWordListGenerator

g2p = PyniniWordListGenerator(
    word_list_path="/path/to/wordlist.txt",
    g2p_model_path="/path/to/english_us_mfa.zip",
    num_pronunciations=2,
)
try:
    g2p.setup()
    g2p.export_pronunciations("/path/to/output.txt")
finally:
    g2p.cleanup()
```

--------------------------------

### Train Acoustic Model CLI

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/workflows/train_acoustic_model.rst

The command-line interface for training acoustic models.

```APIDOC
## mfa train acoustic_model

### Description
Trains an acoustic model using the specified configuration.

### Command
```bash
mfa train <config_path> [--clean] [--overwrite] [--use_mp] [--num_jobs NUM_JOBS]
```

### Parameters
*   **config_path** (string) - Required - Path to the configuration file for training.
*   **--clean** - Optional - Clean intermediate files after training.
*   **--overwrite** - Optional - Overwrite existing models if they exist.
*   **--use_mp** - Optional - Use multiprocessing for training.
*   **--num_jobs** (int) - Optional - Number of jobs to use for multiprocessing.
```

--------------------------------

### Configure MFA Global Settings

Source: https://context7.com/montrealcorpustools/montreal-forced-aligner/llms.txt

Set persistent defaults for MFA commands using the `configure` command. Settings are stored in `~/Documents/MFA/global_config.yaml` and can be overridden per-run.

```bash
# Set number of parallel jobs to 8 and always clean temp files
mfa configure --num_jobs 8 --always_clean

# Set a custom temporary directory and enable PostgreSQL backend
mfa configure --temporary_directory /scratch/mfa_tmp --enable_use_postgres

# Use a named profile (e.g. for multi-project setups)
mfa configure --profile production --always_overwrite --never_debug

# Set GitHub token for faster model downloads (avoids rate limiting)
mfa configure --github_token ghp_XXXXXXXXXXXXXXXX

# View current config by inspecting the yaml file directly
cat ~/Documents/MFA/global_config.yaml
# Expected keys: num_jobs, temporary_directory, clean, use_postgres, verbose, ...
```

--------------------------------

### Customize Phone Topologies

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/user_guide/concepts/hmm.md

Define custom minimum and maximum states for phones in a topology configuration file. This allows for variable modeling of phone durations based on articulation complexity.

```yaml
tʃ:
  - min_states: 3
  - max_states: 5
ɾ:
  - min_states: 1
  - max_states: 1
```

--------------------------------

### Adapt Japanese Acoustic Model

Source: https://github.com/montrealcorpustools/montreal-forced-aligner/blob/main/docs/source/first_steps/alignment_example.md

Adapt a pretrained Japanese acoustic model. This command requires specifying the G2P model path. Adapting the model can lead to better alignment results for Japanese data.

```bash
mfa adapt ~/mfa_data/japanese-jvs-demo-1.0.0 japanese_mfa japanese_mfa ~/mfa_data/japanese_mfa_adapted.zip --g2p_model_path japanese_mfa --clean
```