openai /whisper-large-v3-turbo
Whisper is a state-of-the-art model for automatic speech recognition (ASR) and speech translation, proposed in the paper "Robust Speech Recognition via Large-Scale Weak Supervision" by Alec Radford et al. from OpenAI. Trained on >5M hours of labeled data, Whisper demonstrates a strong ability to generalise to many datasets and domains in a zero-shot setting. Whisper large-v3-turbo is a finetuned version of a pruned Whisper large-v3. In other words, it's the exact same model, except that the number of decoding layers have reduced from 32 to 4. As a result, the model is way faster, at the expense of a minor quality degradation.
By: OpenAI · Released: October 2024 · License: MIT
Core Specs
| Property | Value |
|---|---|
| Parameters | 809M |
| Base Model | Whisper large-v3 (pruned + fine-tuned) |
| Encoder Layers | 32 (unchanged) |
| Decoder Layers | 4 (pruned from 32) |
| Mel Spectrogram Bins | 128 |
| Languages | 99+ |
| Input | Audio |
| Output | Text (transcription) |
| Max Audio Chunk | 30 seconds |
| VRAM | ~6 GB |
| Model File Size | ~1.6 GB |
Architecture
Same encoder-decoder Transformer as large-v3. The only change: decoder pruned from 32 → 4 layers, then fine-tuned via knowledge distillation to recover accuracy.
Encoder handles audio understanding (heavy lifting) — untouched.
Decoder handles token generation — reduced to 4 layers (same astiny).
⚠️ Translation task excluded from Turbo's fine-tuning data — use large-v3 if you need speech translation.
Performance vs. Family
| Model | Params | Decoder Layers | VRAM | Speed |
|---|---|---|---|---|
| tiny | 39M | 4 | ~1 GB | fastest |
| base | 74M | 6 | ~1 GB | — |
| medium | 769M | 24 | ~5 GB | — |
| large-v3-turbo | 809M | 4 | ~6 GB | ~8× faster than large-v3 |
| large-v3 | 1.55B | 32 | ~10 GB | baseline |
Word Error Rate: ~12% — retains ~99% of large-v3 accuracy.
Quick Start
from transformers import pipeline
pipe = pipeline(
"automatic-speech-recognition",
model="openai/whisper-large-v3-turbo",
device="cuda"
)
result = pipe("audio.mp3")
print(result["text"])Hosted APIs
| Provider | Notes |
|---|---|
| Groq | Fastest hosted inference |
| Cloudflare Workers AI | @cf/openai/whisper-large-v3-turbo |
| Hugging Face | openai/whisper-large-v3-turbo |
| OpenRouter | openai/whisper-large-v3-turbo |
