### Install Python Dependencies

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Sets up a virtual environment and installs required Python packages from requirements.txt.

```bash
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
```

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

### Install dpdfnet Package

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Install the dpdfnet package using pip. This command installs the core package for CPU-only inference.

```bash
pip install dpdfnet
```

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

### Download Models from Hugging Face

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Installs the huggingface_hub CLI and downloads PyTorch, ONNX, and TFLite models to local directories.

```bash
pip install -U "huggingface_hub[cli]"

# create target dirs
mkdir -p model_zoo/{checkpoints,onnx,tflite}

# PyTorch checkpoints (HF path: checkpoints/* -> local: model_zoo/checkpoints/*)
hf download Ceva-IP/DPDFNet \
  --include "checkpoints/*.pth" \
  --local-dir model_zoo \

# ONNX models (HF path: onnx/* -> local: model_zoo/onnx/*)
hf download Ceva-IP/DPDFNet \
  --include "onnx/*.onnx" \
  --local-dir model_zoo \

# TFLite models (HF path: *.tflite at repo root -> local: model_zoo/tflite/*)
hf download Ceva-IP/DPDFNet \
  --include "*.tflite" \
  --local-dir model_zoo/tflite \

```

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

### Install dpdfnet with MP3 Support

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Install dpdfnet with optional MP3 support, which requires additional dependencies and FFmpeg.

```bash
pip install 'dpdfnet[mp3]'
```

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

### Install Sounddevice for Real-time Enhancement

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Installs the 'sounddevice' library, which is required for real-time microphone processing with DPDFNet but not included in its dependencies.

```bash
pip install sounddevice
```

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

### Run DNSMOS Local Inference

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Utilize the official DNSMOS local inference script from the DNS Challenge repository for P.835 & P.808 metrics. Ensure you follow their installation and model download instructions.

```python
dnsmos_local.py
```

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

### Predict NISQA v2 Scores

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Use the official NISQA project for NISQA v2 predictions. Refer to their README for environment setup, model weights, and inference commands like running `nisqa_predict.py` on a folder of WAVs.

```python
nisqa_predict.py
```

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

### Enhance Single File with Python API

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Use the dpdfnet Python API to enhance a single audio file and get the output path.

```python
# Enhance one file:
out_path = dpdfnet.enhance_file("noisy.wav", model="dpdfnet2", attn_limit_db=12)
print(out_path)
```

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

### Real-time Microphone Enhancement with DPDFNet StreamEnhancer

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Processes audio from a microphone in real-time using `StreamEnhancer`. This class preserves RNN state across processing calls. The `callback` function handles incoming audio chunks, processes them, and outputs enhanced audio. Ensure `sounddevice` is installed.

```python
import numpy as np
import sounddevice as sd
import dpdfnet

INPUT_SR   = 48000
# Use one model hop (10 ms) as the block size so process() returns
# exactly one hop's worth of enhanced audio on every callback.
BLOCK_SIZE = int(INPUT_SR * 0.010)   # 480 samples at 48 kHz

enhancer = dpdfnet.StreamEnhancer(model="dpdfnet2_48khz_hr")

def callback(indata, outdata, frames, time, status):
    mono_in = indata[:, 0] if indata.ndim > 1 else indata.ravel()
    enhanced = enhancer.process(mono_in, sample_rate=INPUT_SR)
    n = min(len(enhanced), frames)
    outdata[:n, 0] = enhanced[:n]
    if n < frames:
        outdata[n:] = 0.0   # silence while the first window accumulates

with sd.Stream(
    samplerate=INPUT_SR,
    blocksize=BLOCK_SIZE,
    channels=1,
    dtype="float32",
    callback=callback,
):
    print("Enhancing microphone input - press Ctrl+C to stop")
    try:
        while True:
            sd.sleep(100)
    except KeyboardInterrupt:
        pass

# Optional: drain the final partial window at the end of a recording
tail = enhancer.flush()
```

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

### Run Real-Time DPDFNet Demo

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Execute the real-time demo script from the command line. Ensure you are in the repository root and have activated your virtual environment.

```bash
python -m real_time_demo
```

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

### Download All Models

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Download all available speech enhancement models for dpdfnet.

```bash
dpdfnet download
```

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

### Available Models

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Lists all available enhancement models. Iterates through the models and prints their name, readiness status, and whether they are cached.

```APIDOC
## dpdfnet.available_models

### Description
Retrieves a list of available enhancement models, including their status.

### Returns
- **list** - A list of dictionaries, where each dictionary contains 'name', 'ready', and 'cached' keys for a model.

### Request Example
```python
import dpdfnet

for row in dpdfnet.available_models():
    print(row["name"], row["ready"], row["cached"])
```
```

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

### Enhance Directory with CLI (Default Workers)

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Use the dpdfnet CLI to enhance all WAV files in a directory. It uses all available CPU cores by default.

```bash
# Enhance a directory (uses all CPU cores by default)
dpdfnet enhance-dir ./noisy_wavs ./enhanced_wavs --model dpdfnet2 --attn-limit-db 12
```

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

### Download Specific Model with Force Option

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Download a specific speech enhancement model, forcing the download even if it already exists locally.

```bash
dpdfnet download dpdfnet4 --force
```

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

### Display dpdfnet CLI Help

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Show the help message for the dpdfnet command-line interface to understand available commands and options.

```bash
dpdfnet --help
```

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

### Download dpdfnet Models with CLI

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Use the dpdfnet CLI to download available audio enhancement models. You can download all models, a specific model, or force a download.

```bash
# Download models
dpdfnet download
dpdfnet download dpdfnet8
dpdfnet download dpdfnet4 --force
```

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

### Enhance Directory with CLI (Fixed Workers)

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Use the dpdfnet CLI to enhance WAV files in a directory with a specified number of worker processes.

```bash
# Enhance a directory with a fixed worker count
dpdfnet enhance-dir ./noisy_wavs ./enhanced_wavs --model dpdfnet2 --workers 4 --attn-limit-db 12
```

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

### List Available DPDFNet Models

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Retrieves and prints information about available DPDFNet models. The output includes the model name, readiness status, and caching status.

```python
import dpdfnet

for row in dpdfnet.available_models():
    print(row["name"], row["ready"], row["cached"])
```

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

### Calculate PESQ, STOI, SI-SNR

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Use this script to compute PESQ, STOI, and SI-SNR for reference and enhanced audio. It includes auto-alignment for fair comparisons.

```python
pesq_stoi_sisnr_calc.py
```

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

### Enhance Directory of Audio Files (Default Workers)

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Enhance all supported audio files within a directory using the 'enhance-dir' command. This utilizes all available CPU cores by default.

```bash
dpdfnet enhance-dir ./noisy_wavs ./enhanced_wavs --model dpdfnet2 --attn-limit-db 12
```

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

### Download Specific Model

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Download a specific speech enhancement model by its name.

```bash
dpdfnet download dpdfnet8
```

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

### Enhance Single Audio File with CLI

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Use the dpdfnet CLI to enhance a single WAV file. Specify the model and attention limit in decibels.

```bash
# Enhance one file
dpdfnet enhance noisy.wav enhanced.wav --model dpdfnet4 --attn-limit-db 12
```

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

### Enhance Audio In Memory with DPDFNet

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Enhances audio data loaded into memory. Requires 'soundfile' for reading/writing audio and 'dpdfnet' for enhancement. Specify the audio data, sample rate, model, and optionally attention limit in decibels.

```python
import soundfile as sf
import dpdfnet

audio, sr = sf.read("noisy.wav")
enhanced = dpdfnet.enhance(audio, sample_rate=sr, model="dpdfnet4", attn_limit_db=12)
sf.write("enhanced.wav", enhanced, sr)
```

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

### Run Offline Enhancement with ONNX Model

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Performs audio enhancement using an ONNX model. Place noisy WAV files in './noisy_wavs' and specify model details.

```bash
python -m onnx_model.infer_dpdfnet_onnx \
  	--noisy_dir ./noisy_wavs \
	--enhanced_dir ./enhanced_wavs \
	--model_name dpdfnet4 \
	--workers 5 \
	--attn-limit-db 12
```

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

### Download DPDFNet Models

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Downloads DPDFNet models. Call without arguments to download all available models, or specify a model name to download a particular one.

```python
import dpdfnet

dpdfnet.download()
dpdfnet.download("dpdfnet4")
```

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

### Download Models

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Downloads enhancement models. Can download all available models or a specific model by name.

```APIDOC
## dpdfnet.download

### Description
Downloads enhancement models from the API.

### Parameters
- **model_name** (str, optional) - The name of a specific model to download. If not provided, all available models are downloaded.

### Request Example
```python
import dpdfnet

# Download all models
dpdfnet.download()

# Download a specific model
dpdfnet.download("dpdfnet4")
```
```

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

### List Available Models with Python API

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Retrieve a list of available dpdfnet models and their status (ready, cached) using the Python API.

```python
# Model listing:
for row in dpdfnet.available_models():
    print(row["name"], row["ready"], row["cached"])
```

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

### In-Memory Audio Enhancement with Python API

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Enhance audio data directly in memory using the dpdfnet Python API. Requires the soundfile library for reading and writing audio.

```python
import soundfile as sf
import dpdfnet

# In-memory enhancement:
audio, sr = sf.read("noisy.wav")
enhanced = dpdfnet.enhance(audio, sample_rate=sr, model="dpdfnet4", attn_limit_db=12)
sf.write("enhanced.wav", enhanced, sr)
```

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

### Enhance Directory with Fixed Worker Count

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Enhance audio files in a directory with a specified number of worker threads using the 'enhance-dir' command.

```bash
dpdfnet enhance-dir ./noisy_wavs ./enhanced_wavs --model dpdfnet2 --workers 4 --attn-limit-db 12
```

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

### Benchmark ONNX Runtime Performance

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Benchmark the ONNX runtime for DPDFNet to assess its performance and compare its Real-Time Factor (RTF). This is useful for CPU performance troubleshooting.

```python
python -m onnx_model.infer_dpdfnet_onnx --noisy_dir /path/to/noisy/audio --enhanced_dir /path/to/enhanced/audio
```

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

### Enhance File

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Enhances a single audio file. Takes the path to an input audio file, enhances it using a specified model and parameters, and returns the path to the output enhanced file.

```APIDOC
## dpdfnet.enhance_file

### Description
Enhances a single audio file, saving the output to a new file.

### Parameters
- **input_path** (str) - The path to the input audio file.
- **model** (str) - The name of the model to use for enhancement.
- **attn_limit_db** (float, optional) - The attention limit in decibels. Defaults to None.

### Returns
- **str** - The path to the enhanced output audio file.

### Request Example
```python
import dpdfnet

out_path = dpdfnet.enhance_file("noisy.wav", model="dpdfnet2", attn_limit_db=12)
print(out_path)
```
```

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

### In-memory Enhancement

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Enhances an audio array in memory. Reads audio from a file, enhances it using a specified model and parameters, and writes the enhanced audio to a new file.

```APIDOC
## dpdfnet.enhance

### Description
Enhances an audio array in memory using a specified model and parameters.

### Parameters
- **audio** (numpy.ndarray) - The input audio data.
- **sample_rate** (int) - The sample rate of the audio.
- **model** (str) - The name of the model to use for enhancement.
- **attn_limit_db** (float, optional) - The attention limit in decibels. Defaults to None.

### Request Example
```python
import soundfile as sf
import dpdfnet

audio, sr = sf.read("noisy.wav")
enhanced = dpdfnet.enhance(audio, sample_rate=sr, model="dpdfnet4", attn_limit_db=12)
sf.write("enhanced.wav", enhanced, sr)
```
```

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

### Enhance Single Audio File

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Enhance a single audio file using the 'enhance' command. Specify the input file, output file, model, and optional attention limit.

```bash
dpdfnet enhance noisy.wav enhanced.wav --model dpdfnet4 --attn-limit-db 12
```

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

### Run Offline Enhancement with TFLite Model

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Performs audio enhancement using a TFLite model. Place noisy WAV files in './noisy_wavs' and specify model details.

```bash
python -m tflite_model.infer_dpdfnet_tflite \
	--noisy_dir ./noisy_wavs \
	--enhanced_dir ./enhanced_wavs \
	--model_name dpdfnet4 \
	--workers 5 \
	--attn-limit-db 12
```

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

### Enhanced Audio File Naming Convention

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

Describes the naming format for enhanced audio files generated by the model.

```text
<original_stem>_<model_name>.wav
```

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

### Enhance Audio File with DPDFNet

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Enhances a single audio file directly. Specify the input file path, model, and optionally attention limit in decibels. The function returns the path to the enhanced output file.

```python
import dpdfnet

out_path = dpdfnet.enhance_file("noisy.wav", model="dpdfnet2", attn_limit_db=12)
print(out_path)
```

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

### DPDFNet Citation

Source: https://github.com/ceva-ip/dpdfnet/blob/main/README.md

BibTeX entry for citing the DPDFNet paper.

```bibtex
@article{rika2025dpdfnet,
 title = {DPDFNet: Boosting DeepFilterNet2 via Dual-Path RNN},
 author = {Rika, Daniel and Sapir, Nino and Gus, Ido},
 year = {2025},
}
```

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

### StreamEnhancer

Source: https://github.com/ceva-ip/dpdfnet/blob/main/package/README.md

Processes audio chunk-by-chunk for real-time enhancement, preserving RNN state across calls. Suitable for microphone input.

```APIDOC
## dpdfnet.StreamEnhancer

### Description
Processes audio in real-time, maintaining RNN state for continuous enhancement.

### Initialization
- **model** (str) - The name of the model to use for enhancement.

### Methods
- **process(audio_chunk, sample_rate)**: Processes a chunk of audio data.
  - **audio_chunk** (numpy.ndarray): The input audio data chunk.
  - **sample_rate** (int): The sample rate of the audio.
  - **Returns**: numpy.ndarray - The enhanced audio data chunk.
- **flush()**: Processes any remaining buffered audio data.
  - **Returns**: numpy.ndarray - The remaining enhanced audio data.
- **reset()**: Clears the RNN state, useful for processing independent audio segments.

### Notes
- Latency: The first enhanced output arrives after one full model window (~20 ms).
- Sample Rate: `StreamEnhancer` resamples internally. Pass your device's native rate.
- Block Size: Using a block size of one model hop (e.g., 10 ms) yields one enhanced block per callback.

### Real-time Microphone Enhancement Example
```python
import numpy as np
import sounddevice as sd
import dpdfnet

INPUT_SR   = 48000
BLOCK_SIZE = int(INPUT_SR * 0.010) # 480 samples at 48 kHz

enhancer = dpdfnet.StreamEnhancer(model="dpdfnet2_48khz_hr")

def callback(indata, outdata, frames, time, status):
    mono_in = indata[:, 0] if indata.ndim > 1 else indata.ravel()
    enhanced = enhancer.process(mono_in, sample_rate=INPUT_SR)
    n = min(len(enhanced), frames)
    outdata[:n, 0] = enhanced[:n]
    if n < frames:
        outdata[n:] = 0.0

with sd.Stream(
    samplerate=INPUT_SR,
    blocksize=BLOCK_SIZE,
    channels=1,
    dtype="float32",
    callback=callback,
):
    print("Enhancing microphone input - press Ctrl+C to stop")
    try:
        while True:
            sd.sleep(100)
    except KeyboardInterrupt:
        pass

# Optional: drain the final partial window
tail = enhancer.flush()
```
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.