Cantonese speech-to-text

Quickstart

Passing language skips auto-detection, which improves both accuracy on short clips and latency.

curl https://api.inferenceapis.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $INFERENCE_API_KEY" \
  -F "model=openai/whisper-large-v3" \
  -F "language=yue" \
  -F "file=@recording.mp3" \
  -F "response_format=verbose_json"

import os
from openai import OpenAI

client = OpenAI(base_url="https://api.inferenceapis.com/v1", api_key=os.environ["INFERENCE_API_KEY"])
with open("recording.mp3", "rb") as f:
    r = client.audio.transcriptions.create(model="openai/whisper-large-v3", file=f, language="yue", response_format="verbose_json")
print(r.text)

Cost and speed for Cantonese audio

At $0.002 per minute, one hour of Cantonese audio costs $0.12 and ten hours cost $1.20, on either model. In our measurement Whisper Large v3 ran at 0.047× real time, so a 30-minute recording returns in roughly 1.4 minutes. There is no daily cap on minutes; usage is limited by your prepaid balance.

Related languages: Chinese, Japanese, Korean, Vietnamese, Thai.

Tips for Cantonese

• Use language=yue for Cantonese; Whisper outputs Traditional characters for Cantonese.

Subtitles

Use response_format=srt or vtt to get a subtitle file directly; timestamps come from the model's segment boundaries. For word-level timing use verbose_json with timestamp_granularities[]=word.

FAQ

Does auto-detection work for Cantonese?

Yes — omit language and Whisper detects it from the first 30 seconds. Set it explicitly for short clips or mixed-language audio.

Can it translate Cantonese to English?

Transcription returns Cantonese text. For English output, transcribe first and translate the text with a chat model on the same key.