# Whisper review

> Whisper is OpenAI's general-purpose speech recognition model for multilingual transcription, speech translation into English, and language identification, usable from the command line or Python.

Devtoolwatch Board, the week of Oct 25–31, 2026. Canonical page: https://index.devtoolwatch.com/tools/whisper/

## TL;DR

Whisper is a Transformer sequence-to-sequence speech recognition model from OpenAI, trained on a large dataset of diverse audio. It is for Python developers who need transcription, translation into English, or language detection. Six model sizes trade speed for accuracy, and the code and model weights are released under the MIT License.

## Whisper at a glance

- **Category:** [Voice and realtime](https://index.devtoolwatch.com/categories/voice-and-realtime/)
- **Maintained by:** OpenAI (organization, [@openai](https://github.com/openai) on GitHub)
- **License:** MIT
- **Language:** Python
- **Repository:** [github.com/openai/whisper](https://github.com/openai/whisper)
- **PyPI package:** [openai-whisper](https://pypi.org/project/openai-whisper/) 20250625
- **Latest release:** [v20250625](https://github.com/openai/whisper/releases/tag/v20250625) on GitHub, Jun 26, 2025
- **First published:** Sep 16, 2022

## What Whisper does

- Whisper transcribes speech in audio files such as FLAC, MP3, and WAV in many languages.
- Whisper translates non-English speech into English with the multilingual models and the --task translate option.
- Whisper detects the spoken language through whisper.detect_language() in the Python API.
- Whisper offers six model sizes, from tiny to large plus turbo, with different speed, accuracy, and VRAM needs.
- Whisper provides English-only .en models that tend to perform better for English, especially tiny.en and base.en.
- Whisper exposes lower-level Python access through whisper.decode() and DecodingOptions for custom decoding.

### How it works

Whisper runs locally as a command-line tool or a Python library built on PyTorch. It uses ffmpeg to read audio. The transcribe() method reads the entire file and processes the audio with a sliding 30-second window. It makes autoregressive sequence-to-sequence predictions on each window. A single model handles recognition, translation, and language identification. Special tokens specify the task.

## How to install Whisper

**With pip (latest release)**

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

**From the GitHub repository (latest commit)**

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

**Install ffmpeg on macOS with Homebrew**

```
brew install ffmpeg
```

### Requirements

- Python 3.8 to 3.11 and a recent version of PyTorch.
- The ffmpeg command-line tool installed on your system.
- Rust may be needed if tiktoken has no pre-built wheel for your platform.
- GPU memory from about 1 GB for tiny or base up to about 10 GB for large.

### First steps

1. Install ffmpeg with your system package manager, for example: sudo apt update && sudo apt install ffmpeg on Ubuntu or Debian.
2. Transcribe files with the turbo model: whisper audio.flac audio.mp3 audio.wav --model turbo
3. Translate non-English speech into English: whisper japanese.wav --model medium --language Japanese --task translate
4. Run whisper --help to view all available options.

## Who Whisper is for

- Python developers who want local speech-to-text transcription inside their own code.
- Projects that need to translate non-English speech into English text.
- Applications that must identify the spoken language in audio.
- Developers who want open model weights under the MIT License.

### Limitations

- Whisper's turbo model is not trained for translation and returns the original language even with --task translate.
- Whisper's accuracy varies widely depending on the language.
- Whisper's larger models need significant VRAM, about 10 GB for large and about 6 GB for turbo.
- Whisper has a reported issue where complete silence is hallucinated as an Arabic phrase.

## Usage data (the week of Oct 25–31, 2026)

- Board rank: #150 overall, #8 in Voice and realtime, momentum score 52
- Downloads that week: 1.5M (+4% vs 4-week average)
- New GitHub stars per week (last 12 weeks): Aug 9: 443; Aug 16: 477; Aug 23: 369; Aug 30: 436; Sep 6: 450; Sep 13: 461; Sep 20: 296; Sep 27: 363; Oct 4: 370; Oct 11: 586; Oct 18: 487; Oct 25: 486

## Whisper alternatives

| Rank | Tool | Stars | Downloads/wk |
|---|---|---|---|
| #1 | [faster-whisper](https://index.devtoolwatch.com/tools/faster-whisper/) | 26.4k | 4.4M |
| #2 | [Kokoro](https://index.devtoolwatch.com/tools/kokoro/) | 9.8k | 150.7k |
| #3 | [Chatterbox](https://index.devtoolwatch.com/tools/chatterbox/) | 27.5k | 73.2k |
| #4 | [VoiceMem](https://index.devtoolwatch.com/tools/xzf-thu-voicemem/) | 2.8k | – |
| #5 | [Pipecat](https://index.devtoolwatch.com/tools/pipecat/) | 17k | 348.5k |
| #6 | [WhisperX](https://index.devtoolwatch.com/tools/whisperx/) | 25k | 158.6k |

## Whisper FAQ

### Is Whisper free and open source?

Yes. Whisper's source code is on GitHub (openai/whisper) under the MIT license, so you can use it for free.

### How do I install Whisper?

Run `pip install -U openai-whisper`. The openai-whisper package needs Python 3.8 or newer. The install section above has the other options from the project's README.

### Is Whisper still maintained?

Activity has slowed: the openai/whisper repository was last updated on Aug 31, 2026, and the latest release, v20250625 on GitHub, came out on Jun 26, 2025.

### How popular is Whisper?

Whisper has 111.7k GitHub stars as of Nov 1, 2026. It gained 486 new stars in the week of Oct 25–31, 2026, against a 12-week average of 435 a week. Its main package had 1.5M downloads that week, +4% against its own 4-week average. That made it #8 among voice and realtime AI tools on the Devtoolwatch Board.

### Whisper vs WhisperX: how do they compare?

Whisper and WhisperX are both tracked by Devtoolwatch as voice and realtime AI tools. Whisper has 111.7k GitHub stars and WhisperX has 25k. In the week of Oct 25–31, 2026, Whisper's main package had 1.5M downloads and WhisperX's had 158.6k. WhisperX is a Python speech recognition tool built on Whisper that adds word-level timestamps and speaker diarization. [WhisperX review](https://index.devtoolwatch.com/tools/whisperx/)

### What are the best alternatives to Whisper?

Other voice and realtime AI tools on the Devtoolwatch Board in the week of Oct 25–31, 2026, by momentum: faster-whisper (#1), Kokoro (#2), Chatterbox (#3), VoiceMem (#4), Pipecat (#5). The alternatives table above has their numbers.

### How do I use Whisper to transcribe audio?

Whisper transcribes audio with one command: whisper audio.flac audio.mp3 audio.wav --model turbo. In Python, load a model with whisper.load_model("turbo"), call model.transcribe("audio.mp3"), and read result["text"]. For non-English speech, add the --language option, for example --language Japanese.

### Which Whisper model should I use for translation?

Whisper translation needs a multilingual model: tiny, base, small, medium, or large. Do not use turbo, because it is not trained for translation. The README recommends medium or large for the best translation results.

### What is the Whisper turbo model?

Whisper turbo is an optimized version of large-v3. It transcribes faster with a minimal loss in accuracy. It has 809M parameters, needs about 6 GB of VRAM, and is the default model on the command line.

### Does Whisper need a GPU?

Whisper lists approximate VRAM needs for each model size. These range from about 1 GB for tiny and base to about 10 GB for large. The relative speeds in the README were measured on an A100. Real-world speed varies with language, speaking speed, and hardware.

### Can Whisper detect which language is spoken?

Whisper can detect the spoken language. In Python, call model.detect_language() on a log-Mel spectrogram of the audio. It returns probabilities for each language, and the most likely one is the detected language.

## Sources

1. [openai/whisper on GitHub](https://github.com/openai/whisper)
2. [Whisper README](https://github.com/openai/whisper/blob/main/README.md)
3. [openai-whisper on PyPI](https://pypi.org/project/openai-whisper/)
4. [Release v20250625 (Jun 26, 2025)](https://github.com/openai/whisper/releases/tag/v20250625)
5. [Release v20240930 (Sep 30, 2024)](https://github.com/openai/whisper/releases/tag/v20240930)
6. [Release v20240927 (Sep 27, 2024)](https://github.com/openai/whisper/releases/tag/v20240927)
7. [Hacker News: Complete silence is always hallucinated as "ترجمة نانسي قنقر" in Arabic](https://news.ycombinator.com/item?id=44643564)
8. [Hacker News: OpenAI quietly launched Whisper V2 in a GitHub commit](https://news.ycombinator.com/item?id=33884716)
9. [Hacker News: OpenAI releases Whisper v3, new generation open source ASR model](https://news.ycombinator.com/item?id=38166965)
10. [Devtoolwatch Board, the week of Oct 25–31, 2026](https://index.devtoolwatch.com/)