Montreal Forced Aligner

https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/
Admin
Montreal Forced Aligner is a trainable tool for automatic audio-text alignment using acoustic and...
Tokens:28,449
Snippets:384
Trust Score:8.3
Update:3 months ago
Show doc for...
Tokens:
Raw
### Start MFA Database Server

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/server/index.html

Use this command to start the MFA database server. Various options can be specified to configure the server's behavior, such as the profile, temporary directory, number of jobs, and cleaning options.

```bash
mfa server start [OPTIONS]
```

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

### Phone Mapping YAML Example

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/workflows/remap.html

Example of a phone mapping YAML file for remapping alignments. The format is `SOURCE_PHONE: TARGET_PHONE`. For alignment, each source phone must map to exactly one target 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
...
```

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

### Launch Anchor Annotator

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/corpus_creation/anchor.html

Launch the Anchor Annotator GUI utility using the MFA subcommand. Ensure the anchor subpackage has been installed first.

```bash
mfa anchor

```

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

### Advanced YAML Configuration for MFA

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/index.html

An example of a more complex YAML configuration file for MFA, demonstrating settings for beam width, punctuation, and training parameters.

```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

```

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

### List Acoustic Models

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/models/index.html

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

```bash
mfa model list acoustic
```

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

### Install Anchor Subpackage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/corpus_creation/anchor.html

Install the Anchor subpackage for MFA using conda. This command also installs MFA and its dependencies if they are not already present.

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

```

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

### Example Dictionary Entries

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/dictionary.html

Illustrates sample entries in a pronunciation dictionary, showing orthography and its corresponding pronunciation.

```plaintext
c'est S E
c     S E
c'    S
etait E T E
un    A N

```

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

### MFA Server Management CLI

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/server/index.html

The base command for managing MFA database servers. Use this to access subcommands like init, start, stop, and delete.

```bash
mfa server [OPTIONS] COMMAND [ARGS]...
```

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

### Example Orthographic Transcriptions

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/dictionary.html

Provides sample orthographic transcriptions that will be processed by the aligner.

```plaintext
c'est un c
c'etait un c

```

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

### Setting Beam Width for Alignment

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/index.html

Example of setting the beam width for the `mfa align` command directly on the command line. This overrides default or configuration file settings.

```bash
mfa align ... --beam 1000

```

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

### Example Phone Groups for English

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/implementations/phone_groups.html

This excerpt shows how phonetically similar phones and their variants are grouped together in the default English MFA phone groups. This helps in pooling data for similar sounds.

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

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

### Default Training Configuration File

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/acoustic_modeling.html

This configuration file demonstrates the default 2.0 training regime for acoustic modeling. It outlines available options and their progression through the training process, including parameters for feature extraction and different training stages.

```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

```

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

### Initialize MFA Database Server

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/server/index.html

Command to initialize a new MFA database server. This command allows configuration of various aspects of the initialization process, such as the profile, temporary directory, and multiprocessing settings.

```bash
mfa server init [OPTIONS]
```

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

### mfa adapt Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/workflows/adapt_acoustic_model.html

This is the basic usage of the 'mfa adapt' command. It requires the corpus directory, dictionary path, pretrained acoustic model path, and the desired output model path.

```bash
mfa adapt [OPTIONS] CORPUS_DIRECTORY DICTIONARY_PATH ACOUSTIC_MODEL_PATH
          OUTPUT_MODEL_PATH
```

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

### Normalized Transcript Example

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/dictionary.html

The normalized version of the transcript after MFA processing, indicating how cutoff words are handled.

```text
<cutoff-off> with the <cutoff> <cutoff-cut> cut off

```

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

### Normalized Orthographic Transcriptions

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/dictionary.html

Shows the result of text normalization on the example orthographic transcriptions, including handling of clitics.

```plaintext
c'est un c
c' était un c

```

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

### Default Ivector Training Configuration

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/ivector.html

This configuration file demonstrates the available options for training DUBM and ivector extractors, reflecting the current 2.0 training regime. It specifies feature extraction parameters and training settings for both DUBM and ivector stages.

```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

```

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

### Basic YAML Configuration for Beam Width

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/index.html

A simple YAML configuration file to set the beam width. This can be applied using the `--config_path` argument.

```yaml
beam: 1000

```

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

### Download Pretrained Models

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/models/index.html

Downloads pretrained models (acoustic, dictionary, g2p, ivector, language_model, tokenizer) from the MFA repository. If no model names are specified, it lists all downloadable models of the given type.

```bash
mfa model download [OPTIONS]
                   {acoustic|dictionary|g2p|ivector|language_model|tokenizer}
                   [MODEL_NAME]...
```

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

### English Cot-Caught Merger Phonological Rules

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/implementations/phonological_rules.html

Example of phonological rules for the cot-caught merger in English, demonstrating context-dependent vowel replacements for long and short variants.

```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
    preprevious_context: ''
    replacement: ɑ
    segment: ɒ

```

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

### Download Pretrained Models

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/models/index.html

Downloads pretrained models for aligning and generating pronunciation dictionaries. Specify the model type as 'acoustic', 'g2p', or 'dictionary'.

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

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

### MFA Configure Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/index.html

This shows the basic usage of the `mfa configure` command. Use flags to set default values for MFA operations.

```bash
mfa configure [OPTIONS]

```

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

### Default G2P Training Configuration

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/g2p.html

This configuration file sets parameters for training a G2P model. It includes options for punctuation handling, clitic and compound markers, G2P model order, training randomization, and Baum-Welch algorithm settings.

```yaml
punctuation: "、。।，@<>\"(),.:;¿?¡!\\&%#*~【】，…‥「」『』〝〟″⟨⟩♪・‹›«»～′$+"
clitic_markers: "'’"
compound_markers: "-"
num_pronunciations: 1  # Used if running in validation mode
order: 7
random_starts: 25
seed: 1917
delta: 0.0009765
lr: 1.0
batch_size: 200
max_iterations: 10
smoothing_method: "kneser_ney"
pruning_method: "relative_entropy"
model_size: 1000000

```

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

### Stop MFA Database Server

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/server/index.html

Use this command to stop the MFA database server. You can specify a mode for stopping the server (fast, immediate, smart) and other configuration options similar to starting the server.

```bash
mfa server stop [OPTIONS]
```

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

### Acoustic Model Training Configuration (v1.0)

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/acoustic_modeling.html

This configuration block defines the parameters for training acoustic models, including feature extraction and different training stages like monophone, triphone, and SAT.

```yaml
beam: 10
retry_beam: 40

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

training:
  - monophone:
      num_iterations: 40
      max_gaussians: 1000
      boost_silence: 1.0

  - triphone:
      num_iterations: 35
      num_leaves: 3100
      max_gaussians: 50000
      cluster_threshold: 100
      boost_silence: 1.0
      power: 0.25

  - sat:
      num_leaves: 3100
      max_gaussians: 50000
      power: 0.2
      silence_weight: 0.0
      cluster_threshold: 100
      fmllr_update_type: "full"

```

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

### Train a word tokenizer command usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/corpus_creation/train_tokenizer.html

This is the basic command structure for training a tokenizer model. It requires the path to the tokenized corpus directory and the desired output model path.

```bash
mfa train_tokenizer [OPTIONS] CORPUS_DIRECTORY OUTPUT_MODEL_PATH
```

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

### MFA Train G2P Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/workflows/g2p_train.html

Basic usage of the mfa train_g2p command. Specify the path to the pronunciation dictionary and the desired output model path.

```bash
mfa train_g2p [OPTIONS] DICTIONARY_PATH OUTPUT_MODEL_PATH

```

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

### MFA Command with Per-speaker Dictionary

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/dictionary.html

This command demonstrates how to use a per-speaker dictionary YAML file with the MFA align command. Replace the placeholder paths with your actual file locations.

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

```

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

### List Local Models

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/models/index.html

Lists all locally saved models of a specified type (acoustic, dictionary, g2p, ivector, language_model, tokenizer). This helps in managing and identifying available models on your system.

```bash
mfa model list [OPTIONS]
               {acoustic|dictionary|g2p|ivector|language_model|tokenizer}
```

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

### Generate Pronunciation Dictionary with MFA G2P

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/workflows/dictionary_generating.html

Use this command to generate a pronunciation dictionary using a G2P model. Specify input, G2P model, and output paths.

```bash
mfa g2p [OPTIONS] INPUT_PATH G2P_MODEL_PATH OUTPUT_PATH
```

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

### Customize Phone Topology States

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/concepts/hmm.html

Define minimum and maximum states for phones in a topology configuration file. Use this to control the complexity of phone modeling, accommodating variations in articulation duration.

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

```

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

### Train Dictionary Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/corpus_creation/training_dictionary.html

This snippet shows the basic usage of the 'mfa train_dictionary' command, including the required arguments for corpus directory, dictionary path, acoustic model path, and output directory.

```bash
mfa train_dictionary [OPTIONS] CORPUS_DIRECTORY DICTIONARY_PATH
                     ACOUSTIC_MODEL_PATH OUTPUT_DIRECTORY

```

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

### Default Dictionary Generation Configuration

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/g2p.html

This configuration file specifies parameters for generating pronunciation dictionaries using a G2P model. It includes settings for punctuation and marker handling, and the number of pronunciations to generate.

```yaml
punctuation: "、。।，@<>\"(),.:;¿?¡!\\&%#*~【】，…‥「」『』〝〟″⟨⟩♪・‹›«»～′$+"
clitic_markers: "'’"
compound_markers: "-"
num_pronunciations: 1

```

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

### Tokenize Utterances Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/corpus_creation/tokenize.html

This snippet shows the basic usage of the 'mfa tokenize' command. It requires input audio/text paths, a tokenizer model, and an output directory.

```bash
mfa tokenize [OPTIONS] INPUT_PATH TOKENIZER_MODEL_PATH OUTPUT_DIRECTORY

```

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

### mfa align_one Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/workflows/alignment.html

This is the basic command structure for aligning a single file. Ensure all required file paths and model paths are correctly specified.

```bash
mfa align_one [OPTIONS] SOUND_FILE_PATH TEXT_FILE_PATH DICTIONARY_PATH
              ACOUSTIC_MODEL_PATH OUTPUT_PATH
```

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

### Save Model

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/models/index.html

Saves a model to the pretrained folder for later use. This command requires the path to the model and its type.

```bash
mfa model save [OPTIONS]
               {acoustic|dictionary|g2p|ivector|language_model|tokenizer} PATH
```

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

### Update MFA Software

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/troubleshooting.html

Run this command to update to the latest available versions of MFA. This can resolve issues caused by outdated software.

```bash
mfa_update
```

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

### Inspect Model Metadata

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/models/index.html

Inspects a specified model (acoustic, dictionary, g2p, ivector, language_model, tokenizer) and displays its metadata. This is useful for understanding the properties of a downloaded or local model.

```bash
mfa model inspect [OPTIONS]
                  {acoustic|dictionary|g2p|ivector|language_model|tokenizer}
                  MODEL
```

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

### Train Language Model Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/corpus_creation/training_lm.html

Basic usage of the mfa train_lm command. Specify the source path for the corpus and the output path for the trained language model.

```bash
mfa train_lm [OPTIONS] SOURCE_PATH OUTPUT_MODEL_PATH
```

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

### Train Ivector Extractor Command

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/corpus_creation/train_ivector.html

This command trains an ivector extractor from a corpus and a pretrained acoustic model. Specify the corpus directory and the output model path as arguments. Various options are available for configuration, including paths for config files, audio directories, temporary directories, and multiprocessing settings.

```bash
mfa train_ivector [OPTIONS] CORPUS_DIRECTORY OUTPUT_MODEL_PATH
```

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

### Download Default English Acoustic Model

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/models/index.html

Downloads the default English acoustic model trained on LibriSpeech. After this, `mfa model list acoustic` will show 'english_us_arpa'.

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

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

### Default Segmentation Configuration

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/segment.html

This is the default configuration file used for audio segmentation. It sets parameters like energy threshold, mean scale, maximum segment length, and minimum pause duration.

```yaml
energy_threshold: 5.5
energy_mean_scale: 0.5
max_segment_length: 30
min_pause_duration: 0.05

```

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

### Overriding YAML Configuration with Command Line Argument

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/configuration/index.html

Demonstrates how a command-line argument (`--beam`) takes precedence over a setting in a specified YAML configuration file.

```bash
mfa align ... --beam 10 --config_path config_above.yaml

```

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

### MFA Train Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/workflows/train_acoustic_model.html

Basic usage of the mfa train command, specifying the corpus directory, dictionary path, and output model path.

```bash
mfa train [OPTIONS] CORPUS_DIRECTORY DICTIONARY_PATH OUTPUT_MODEL_PATH
```

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

### Remap Dictionary Command Usage

Source: https://montreal-forced-aligner.readthedocs.io/en/latest/user_guide/workflows/remap.html

This is the basic usage of the 'mfa remap dictionary' command. It requires paths to the input dictionary, acoustic model, phone mapping, and the desired output dictionary.

```bash
mfa remap dictionary [OPTIONS] DICTIONARY_PATH ACOUSTIC_MODEL_PATH
                     PHONE_MAPPING_PATH OUTPUT_DICTIONARY_PATH

```
Tokens:
Raw