Whisper

[Blog] [Paper] [Model card] [Colab example]

Whisper is a general-purpose speech recognition model. It is trained on a large dataset of diverse audio and is also a multi-task model that can perform multilingual speech recognition as well as speech translation and language identification.

Approach

A Transformer sequence-to-sequence model is trained on various speech processing tasks, including multilingual speech recognition, speech translation, spoken language identification, and voice activity detection. All of these tasks are jointly represented as a sequence of tokens to be predicted by the decoder, allowing for a single model to replace many different stages of a traditional speech processing pipeline. The multitask training format uses a set of special tokens that serve as task specifiers or classification targets.

Setup

We used Python 3.9.9 and PyTorch 1.10.1 to train and test our models, but the codebase is expected to be compatible with Python 3.7 or later and recent PyTorch versions. The codebase also depends on a few Python packages, most notably HuggingFace Transformers for their fast tokenizer implementation and ffmpeg-python for reading audio files. The following command will pull and install the latest commit from this repository, along with its Python dependencies

pip install git+https://github.com/openai/whisper.git 

It also requires the command-line tool ffmpeg to be installed on your system, which is available from most package managers:

# on Ubuntu or Debian
sudo apt update && sudo apt install ffmpeg

# on MacOS using Homebrew (https://brew.sh/)
brew install ffmpeg                         

# on Windows using Chocolatey (https://chocolatey.org/)
choco install ffmpeg

Available models and languages

There are five model sizes, four with English-only versions, offering speed and accuracy tradeoffs. Below are the names of the available models and their approximate memory requirements and relative speed.

Size	Parameters	English-only model	Multilingual model	Required VRAM	Relative speed
tiny	39 M	tiny.en	tiny	~1 GB	~32x
base	74 M	base.en	base	~1 GB	~16x
small	244 M	small.en	small	~2 GB	~6x
medium	769 M	medium.en	medium	~5 GB	~2x
large	1550 M	N/A	large	~10 GB	1x

For English-only applications, the .en models tend to perform better, especially for the tiny.en and base.en models. We observed that the difference becomes less significant for the small.en and medium.en models.

Whisper's performance varies widely depending on the language. The figure below shows a WER breakdown by languages of Fleurs dataset, using the large model. More WER and BLEU scores corresponding to the other models and datasets can be found in Appendix D in the paper.

Command-line usage

The following command will transcribe speech in audio files, using the medium model:

whisper audio.flac audio.mp3 audio.wav --model medium

The default setting (which selects the small model) works well for transcribing English. To transcribe an audio file containing non-English speech, you can specify the language using the --language option:

whisper japanese.wav --language Japanese

Adding --task translate will translate the speech into English:

whisper japanese.wav --language Japanese --task translate

Run the following to view all available options:

whisper --help

See tokenizer.py for the list of all available languages.

Python usage

Transcription can also be performed within Python:

import whisper

model = whisper.load_model("base")
result = model.transcribe("audio.mp3")
print(result["text"])

Internally, the transcribe() method reads the entire file and processes the audio with a sliding 30-second window, performing autoregressive sequence-to-sequence predictions on each window.

Below is an example usage of whisper.detect_language() and whisper.decode() which provide lower-level access to the model.

import whisper

model = whisper.load_model("base")

# load audio and pad/trim it to fit 30 seconds
audio = whisper.load_audio("audio.mp3")
audio = whisper.pad_or_trim(audio)

# make log-Mel spectrogram and move to the same device as the model
mel = whisper.log_mel_spectrogram(audio).to(model.device)

# detect the spoken language
_, probs = model.detect_language(mel)
print(f"Detected language: {max(probs, key=probs.get)}")

# decode the audio
options = whisper.DecodingOptions()
result = whisper.decode(model, mel, options)

# print the recognized text
print(result.text)

License

The code and the model weights of Whisper are released under the MIT License. See LICENSE for further details.