Whisper Large v3 - ONNX

Official OpenAI Whisper large-v3 checkpoint exported as ONNX encoder/decoder graphs for speech-core and sherpa-onnx-compatible runtimes.

Part of the soniqo.audio speech toolkit. This is the ONNX Runtime bundle used by speech-core for server, desktop, and Android-style runtimes; the graphs remain sherpa-onnx compatible. Browse ONNX bundles in the soniqo ONNX collection.

Model


Source	openai/whisper-large-v3
Export format	ONNX for sherpa-onnx
Variants	FP32, INT8, FP16
Runtime	speech-core `OnnxWhisperStt` / ONNX Runtime; sherpa-onnx compatible
Total artifact size	10531.36 MiB

The FP16 graphs use external *.onnx.data files. Large-style bundles also use external FP32 *.weights files. Keep external-data files beside their matching .onnx files.

Files

File	Size	Description
`large-v3-decoder.fp16.onnx`	2.23 MB	Whisper decoder graph
`large-v3-decoder.fp16.onnx.data`	1528.97 MB	External tensor data for the adjacent ONNX graph
`large-v3-decoder.int8.onnx`	961.49 MB	Whisper decoder graph
`large-v3-decoder.onnx`	2.6 MB	Whisper decoder graph
`large-v3-decoder.weights`	3057.95 MB	External tensor data for the adjacent ONNX graph
`large-v3-encoder.fp16.onnx`	0.64 MB	Whisper encoder graph
`large-v3-encoder.fp16.onnx.data`	1415.0 MB	External tensor data for the adjacent ONNX graph
`large-v3-encoder.int8.onnx`	731.08 MB	Whisper encoder graph
`large-v3-encoder.onnx`	0.63 MB	Whisper encoder graph
`large-v3-encoder.weights`	2830.0 MB	External tensor data for the adjacent ONNX graph
`large-v3-tokens.txt`	0.78 MB	Tokenizer tokens for speech-core and sherpa-onnx-compatible runtimes

Usage

Use with speech-core's native ONNX Whisper runtime:

#include <speech_core/models/onnx_whisper_stt.h>

speech_core::OnnxWhisperStt stt(
    "large-v3-encoder.int8.onnx",
    "large-v3-decoder.int8.onnx",
    "large-v3-tokens.txt");

auto result = stt.transcribe(audio, length, 16000);

The same encoder/decoder/token files can also be loaded by sherpa-onnx:

import sherpa_onnx

recognizer = sherpa_onnx.OfflineRecognizer.from_whisper(
    encoder="large-v3-encoder.fp16.onnx",
    decoder="large-v3-decoder.fp16.onnx",
    tokens="large-v3-tokens.txt",
    language="en",
    task="transcribe",
    provider="cpu",
)

Benchmarks

This repository ships the exported model artifacts. Full published WER/RTF is currently available for the turbo bundle; run the benchmark command above for local numbers on this variant.

soniqo/Whisper-Large-v3-ONNX