---
license: mit
language: it
library_name: transformers
pipeline_tag: automatic-speech-recognition
thumbnail: null
tags:
- automatic-speech-recognition
- hf-asr-leaderboard
datasets:
- mozilla-foundation/common_voice_17_0
- facebook/multilingual_librispeech
- facebook/voxpopuli
- espnet/yodas
metrics:
- wer
---

# Whisper-Large-V3-Distil-Italian-v0.2

A distilled version of Whisper with 2 decoder layers, optimized for Italian speech-to-text.

This version extends the training to 30-second audio segments to maintain long-form transcription abilities. The training process used a ["patient" teacher](https://arxiv.org/abs/2106.05237) during distillation - meaning longer training times and more aggressive data augmentation - which improved overall performance.

The model uses [openai/whisper-large-v3](https://huggingface.co/openai/whisper-large-v3) as the teacher model while keeping the encoder architecture unchanged. This makes it suitable as a draft model for speculative decoding, potentially getting 2x inference speed while maintaining identical outputs by only adding 2 extra decoder layers and running the encoder just once. It can also serve as a standalone model to trade some accuracy for better efficiency, running 5.8x faster while using only 49% of the parameters. This [paper](https://arxiv.org/abs/2311.00430) also suggests that the distilled model may actually produce fewer hallucinations than the full model during long-form transcription.

The model has been converted into multiple formats to ensure broad compatibility across libraries including transformers, openai-whisper, faster-whisper, whisper.cpp, candle, mlx.

## Performance

The model was evaluated on both short and long-form transcriptions, using in-distribution (ID) and out-of-distribution (OOD) datasets to assess accuracy, generalizability, and robustness.

Note that Word Error Rate (WER) results shown here are [post-normalization](https://github.com/openai/whisper/blob/main/whisper/normalizers/basic.py), which includes converting text to lowercase and removing symbols and punctuation.

All evaluation results on the public datasets can be found [here](https://drive.google.com/drive/folders/1g712e7xvN4SzGdzxtf4OrpT8raWGz_wR?usp=drive_link).

### Short-Form Transcription

![eval-short-form](https://huggingface.co/bofenghuang/whisper-large-v3-distil-it-v0.2/resolve/main/assets/eval_short_form.png)

*Italic* indicates in-distribution (ID) evaluation, where test sets correspond to data distributions seen during training, typically yielding higher performance than out-of-distribution (OOD) evaluation. *~~Italic and strikethrough~~* denotes potential test set contamination - for example, when training and evaluation use different versions of Common Voice, raising the possibility of overlapping data.

### Long-Form Transcription

Long-form transcription evaluation used the 🤗 Hugging Face [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline) with both [chunked](https://huggingface.co/blog/asr-chunking) (chunk_length_s=30) and original sequential decoding methods.

![eval-long-form](https://huggingface.co/bofenghuang/whisper-large-v3-distil-it-v0.2/resolve/main/assets/eval_long_form.png)

## Usage

### Hugging Face Pipeline

The model can be easily used with the 🤗 Hugging Face [`pipeline`](https://huggingface.co/docs/transformers/main_classes/pipelines#transformers.AutomaticSpeechRecognitionPipeline) class for audio transcription. For long-form transcription (over 30 seconds), it will perform sequential decoding as described in OpenAI's paper. If you need faster inference, you can use the `chunk_length_s` argument for [chunked parallel decoding](https://huggingface.co/blog/asr-chunking), which provides 9x faster inference speed but may slightly compromise performance compared to OpenAI's sequential algorithm.

```python
import torch
from datasets import load_dataset
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline

device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Load model
model_name_or_path = "bofenghuang/whisper-large-v3-distil-it-v0.2"
processor = AutoProcessor.from_pretrained(model_name_or_path)
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_name_or_path,
    torch_dtype=torch_dtype,
    low_cpu_mem_usage=True,
)
model.to(device)

# Init pipeline
pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    feature_extractor=processor.feature_extractor,
    tokenizer=processor.tokenizer,
    torch_dtype=torch_dtype,
    device=device,
    # chunk_length_s=30,  # for chunked decoding
    max_new_tokens=128,
)

# Example audio
dataset = load_dataset("bofenghuang/asr-dummy", "it", split="test")
sample = dataset[0]["audio"]

# Run pipeline
result = pipe(sample)
print(result["text"])
```

### Hugging Face Low-level APIs

You can also use the 🤗 Hugging Face low-level APIs for transcription, offering greater control over the process, as demonstrated below:

```python
import torch
from datasets import load_dataset
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor

device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Load model
model_name_or_path = "bofenghuang/whisper-large-v3-distil-it-v0.2"
processor = AutoProcessor.from_pretrained(model_name_or_path)
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_name_or_path,
    torch_dtype=torch_dtype,
    low_cpu_mem_usage=True,
)
model.to(device)

# Example audio
dataset = load_dataset("bofenghuang/asr-dummy", "it", split="test")
sample = dataset[0]["audio"]

# Extract feautres
input_features = processor(
    sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt"
).input_features


# Generate tokens
predicted_ids = model.generate(
    input_features.to(dtype=torch_dtype).to(device), max_new_tokens=128
)

# Detokenize to text
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)[0]
print(transcription)
```

### Speculative Decoding

[Speculative decoding](https://huggingface.co/blog/whisper-speculative-decoding) can be achieved using a draft model, essentially a distilled version of Whisper. This approach guarantees identical outputs to using the main Whisper model alone, offers a 2x faster inference speed, and incurs only a slight increase in memory overhead. 

Since the distilled Whisper has the same encoder as the original, only its decoder need to be loaded, and encoder outputs are shared between the main and draft models during inference.

Using speculative decoding with the Hugging Face pipeline is simple - just specify the `assistant_model` within the generation configurations.

```python
import torch
from datasets import load_dataset
from transformers import (
    AutoModelForCausalLM,
    AutoModelForSpeechSeq2Seq,
    AutoProcessor,
    pipeline,
)

device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Load model
model_name_or_path = "openai/whisper-large-v3"
processor = AutoProcessor.from_pretrained(model_name_or_path)
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_name_or_path,
    torch_dtype=torch_dtype,
    low_cpu_mem_usage=True,
)
model.to(device)

# Load draft model
assistant_model_name_or_path = "bofenghuang/whisper-large-v3-distil-it-v0.2"
assistant_model = AutoModelForCausalLM.from_pretrained(
    assistant_model_name_or_path,
    torch_dtype=torch_dtype,
    low_cpu_mem_usage=True,
)
assistant_model.to(device)

# Init pipeline
pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    feature_extractor=processor.feature_extractor,
    tokenizer=processor.tokenizer,
    torch_dtype=torch_dtype,
    device=device,
    generate_kwargs={"assistant_model": assistant_model},
    max_new_tokens=128,
)

# Example audio
dataset = load_dataset("bofenghuang/asr-dummy", "it", split="test")
sample = dataset[0]["audio"]

# Run pipeline
result = pipe(sample)
print(result["text"])
```

### OpenAI Whisper

You can also employ the sequential long-form decoding algorithm with a sliding window and temperature fallback, as outlined by OpenAI in their original [paper](https://arxiv.org/abs/2212.04356).

First, install the [openai-whisper](https://github.com/openai/whisper) package:

```bash
pip install -U openai-whisper
```

Then, download the converted model:

```bash
huggingface-cli download --include original_model.pt --local-dir ./models/whisper-large-v3-distil-it-v0.2 bofenghuang/whisper-large-v3-distil-it-v0.2
```

Now, you can transcirbe audio files by following the usage instructions provided in the repository:

```python
import whisper
from datasets import load_dataset

# Load model
model_name_or_path = "./models/whisper-large-v3-distil-it-v0.2/original_model.pt"
model = whisper.load_model(model_name_or_path)

# Example audio
dataset = load_dataset("bofenghuang/asr-dummy", "it", split="test")
sample = dataset[0]["audio"]["array"].astype("float32")

# Transcribe
result = model.transcribe(sample, language="it")
print(result["text"])
```

### Faster Whisper

Faster Whisper is a reimplementation of OpenAI's Whisper models and the sequential long-form decoding algorithm in the [CTranslate2](https://github.com/OpenNMT/CTranslate2) format.

Compared to openai-whisper, it offers up to 4x faster inference speed, while consuming less memory. Additionally, the model can be quantized into int8, further enhancing its efficiency on both CPU and GPU.

First, install the [faster-whisper](https://github.com/SYSTRAN/faster-whisper) package:

```bash
pip install faster-whisper
```

Then, download the model converted to the CTranslate2 format:

```bash
huggingface-cli download --include ctranslate2/* --local-dir ./models/whisper-large-v3-distil-it-v0.2 bofenghuang/whisper-large-v3-distil-it-v0.2
```

Now, you can transcirbe audio files by following the usage instructions provided in the repository:

```python
from datasets import load_dataset
from faster_whisper import WhisperModel

# Load model
model_name_or_path = "./models/whisper-large-v3-distil-it-v0.2/ctranslate2"
model = WhisperModel(model_name_or_path", device="cuda", compute_type="float16")  # Run on GPU with FP16

# Example audio
dataset = load_dataset("bofenghuang/asr-dummy", "it", split="test")
sample = dataset[0]["audio"]["array"].astype("float32")

segments, info = model.transcribe(sample, beam_size=5, language="it")

for segment in segments:
    print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))
```

### Whisper.cpp

Whisper.cpp is a reimplementation of OpenAI's Whisper models, crafted in plain C/C++ without any dependencies. It offers compatibility with various backends and platforms.

Additionally, the model can be quantized to either 4-bit or 5-bit integers, further enhancing its efficiency.

First, clone and build the [whisper.cpp](https://github.com/ggerganov/whisper.cpp) repository:

```bash
git clone https://github.com/ggerganov/whisper.cpp.git
cd whisper.cpp

# build the main example
make
```

Next, download the converted ggml weights from the Hugging Face Hub:

```bash
# Download model quantized with Q5_0 method
huggingface-cli download --include ggml-model* --local-dir ./models/whisper-large-v3-distil-it-v0.2 bofenghuang/whisper-large-v3-distil-it-v0.2
```

Now, you can transcribe an audio file using the following command:

```bash
./main -m ./models/whisper-large-v3-distil-it-v0.2/ggml-model-q5_0.bin -l it -f /path/to/audio/file --print-colors
```

### Candle

[Candle-whisper](https://github.com/huggingface/candle/tree/main/candle-examples/examples/whisper) is a reimplementation of OpenAI's Whisper models in the candle format - a lightweight ML framework built in Rust.

First, clone the [candle](https://github.com/huggingface/candle) repository:

```bash
git clone https://github.com/huggingface/candle.git
cd candle/candle-examples/examples/whisper
```

Transcribe an audio file using the following command:

```bash
cargo run --example whisper --release -- --model large-v3 --model-id bofenghuang/whisper-large-v3-distil-it-v0.2 --language it --input /path/to/audio/file
```

In order to use CUDA add `--features cuda` to the example command line:

```bash
cargo run --example whisper --release --features cuda -- --model large-v3 --model-id bofenghuang/whisper-large-v3-distil-it-v0.2 --language it --input /path/to/audio/file
```

### MLX

[MLX-Whisper](https://github.com/ml-explore/mlx-examples/tree/main/whisper) is a reimplementation of OpenAI's Whisper models in the [MLX](https://github.com/ml-explore/mlx) format - a ML framework on Apple silicon. It supports features like lazy computation, unified memory management, etc.

First, clone the [MLX Examples](https://github.com/ml-explore/mlx-examples) repository:

```bash
git clone https://github.com/ml-explore/mlx-examples.git
cd mlx-examples/whisper
```

Next, install the dependencies:

```bash
pip install -r requirements.txt
```

Download the pytorch checkpoint in the original OpenAI format and convert it into MLX format (We haven't included the converted version here since the repository is already heavy and the conversion is very fast):

```bash
# Download
huggingface-cli download --include original_model.pt --local-dir ./models/whisper-large-v3-distil-it-v0.2 bofenghuang/whisper-large-v3-distil-it-v0.2
# Convert into .npz
python convert.py --torch-name-or-path ./models/whisper-large-v3-distil-it-v0.2/original_model.pt --mlx-path ./mlx_models/whisper-large-v3-distil-it-v0.2
```

Now, you can transcribe audio with:

```python
import whisper

result = whisper.transcribe("/path/to/audio/file", path_or_hf_repo="mlx_models/whisper-large-v3-distil-it-v0.2", language="it")
print(result["text"])
```

## Training details

We built a Italian speech recognition dataset of over 11,000 hours of annotated and semi-annotated speech. After decoding this dataset through Whisper-Large-V3 and filtering out segments with WER above 20%, we retained approximately 6,500 hours of high-quality audio.

| Dataset | Total Duration (h) | Filtered Duration (h) <20% WER |
|---|---:|---:|
| mcv | 249.92 | 232.87 |
| mls | 247.38 | 234.14 |
| voxpopuli | 74.11 | 58.25 |
| mtedx | 94.10 | 88.69 |
| yodas-it000 | 1447.25 | 953.19 |
| yodas-it100 | 4929.73 | 2665.54 |
| yodas-it101 | 4192.61 | 2275.90 |
| total | 11235.10 | 6508.58 |

Most data were first concatenated into 30-second segments, primarily preserving the same speaker, then inferred together. 50% of segments were trained with timestamps to ensure good timestamp prediction, and only 20% of segments were trained with previous context since we don't expect the 2-layer decoder to excel at this task.

This model was trained for a fairly long schedule of 100 epochs using aggressive data augmentation, with eval WER continuing to decrease. Some hyperparameter choices were made to favor long-form over short-form transcription. For further details, please refer to the [Distil-Whisper](https://github.com/huggingface/distil-whisper) repository.

All model training was conducted on the [Jean-Zay supercomputer](http://www.idris.fr/eng/jean-zay/jean-zay-presentation-eng.html) at GENCI. Special thanks to the IDRIS team for their excellent support throughout this project.

## Acknowledgements

- OpenAI for developing and open-sourcing the [Whisper](https://arxiv.org/abs/2212.04356) model
- 🤗 Hugging Face for implementing Whisper in the [Transformers](https://github.com/huggingface/transformers) library and creating [Distil-Whisper](https://github.com/huggingface/distil-whisper)
- [Genci](https://genci.fr/) for generously providing the GPU computing resources for this project