Speech-to-Text

Use Whisper models to convert audio files to text, supporting transcription and translation features

Introduction

The Speech-to-Text (STT) API based on OpenAI’s Whisper models can convert audio files to text. It supports various use cases:

• Transcribing audio files to text
• Translating multilingual audio to English
• Supporting multiple audio format inputs
• Providing multiple output format options

Available Model List:

• whisper-large-v3 —— Latest large Whisper model, supports multiple languages. For Chinese recognition, use with appropriate prompts and low temperature values
• whisper-1 —— Original Whisper model, stable and reliable, supports multiple languages
• distil-whisper-large-v3-en —— Distilled model, faster processing speed but slightly lower accuracy, recommended with low temperature values

Performance Recommendations:

• For Chinese audio, recommend using whisper-large-v3 model with appropriate prompts and lower temperature values (e.g., 0.2) to reduce hallucinations
• For English audio or faster processing, use distil-whisper-large-v3-en model
• Supported audio formats: mp3, mp4, mpeg, mpga, m4a, wav, webm
• File size limit: maximum 25MB

Model Usage

Speech Transcription

Use /v1/audio/transcriptions endpoint via client.audio.transcriptions.create() method to transcribe audio to text in the original language.

Speech Translation

Use /v1/audio/translations endpoint via client.audio.translations.create() method to translate audio to English text.

Request Parameters

Transcription Parameters

file · file · required

Audio file object to transcribe. Supported formats: mp3, mp4, mpeg, mpga, m4a, wav, webm, maximum 25MB

model · string · required

Model ID to use. Options: whisper-large-v3, whisper-1, distil-whisper-large-v3-en

language · string

Language of the input audio in ISO-639-1 format (e.g., ‘en’, ‘zh’). Specifying the language can improve accuracy and latency

prompt · string

Optional text prompt to guide the model’s style or continue a previous audio segment. The prompt should match the audio language

response_format · string

Transcription output format. Options: json (default), text, srt, verbose_json, vtt

temperature · number

Sampling temperature between 0 and 1. Higher values make output more random, lower values make it more focused and deterministic. Default is 0

timestamp_granularities[] · array

Timestamp granularities. Options: word, segment. Only available when response_format is verbose_json

Translation Parameters

file · file · required

Audio file object to translate. Same formats as transcription

model · string · required

Model ID to use, same as transcription parameters

prompt · string

Optional English text prompt to guide translation style

response_format · string

Translation output format, same as transcription parameters

temperature · number

Sampling temperature, same as transcription parameters

Usage Examples

Curl Transcription

curl https://aihubmix.com/v1/audio/transcriptions \
  -H "Authorization: Bearer $AIHUBMIX_API_KEY" \
  -H "Content-Type: multipart/form-data" \
  -F file="@/path/to/file/audio.mp3" \
  -F model="whisper-large-v3" \
  -F response_format="text" \
  -F temperature="0.2"

Response Formats

JSON Format (Default)

{
  "text": "This is the transcribed text content"
}

Verbose JSON Format (verbose_json)

{
  "task": "transcribe",
  "language": "english",
  "duration": 8.470000267028809,
  "text": "This is the transcribed text content",
  "segments": [
    {
      "id": 0,
      "seek": 0,
      "start": 0.0,
      "end": 8.470000267028809,
      "text": " This is the transcribed text content",
      "tokens": [50364, 50365, 50365, 50365],
      "temperature": 0.2,
      "avg_logprob": -0.9929364013671875,
      "compression_ratio": 0.8888888888888888,
      "no_speech_prob": 0.0963134765625
    }
  ]
}

Text Format

This is the transcribed text content

SRT Format

1
00:00:00,000 --> 00:00:08,470
This is the transcribed text content

VTT Format

WEBVTT

00:00:00.000 --> 00:00:08.470
This is the transcribed text content

Best Practices

1. Chinese Audio Processing: Use whisper-large-v3 model, set language="zh", temperature=0.2, and provide appropriate Chinese prompts
2. English Audio Processing: Use distil-whisper-large-v3-en for faster processing speed
3. Noise Handling: Use prompts to instruct the model to ignore background noise or clean up stammering issues
4. Long Audio Processing: API automatically segments long audio; recommend preprocessing audio quality for best results
5. Timestamp Requirements: Use verbose_json format and timestamp_granularities when precise timestamps are needed
6. Subtitle Creation: Use srt or vtt format output directly without additional processing

---

Last updated: 2026-06-01