Typhoon ASR Documentation

About The Model

• Typhoon ASR Real-Time:

  • Full Release Information

  • Model Fact Sheet

  • Web Playground - Try it instantly in your browser. Great for casual users.

🎧 Supported File Types

.wav, .mp3, .flac, .ogg, .opus

🔌 Option 1: Use Typhoon API

Typhoon’s hosted API is OpenAI-compatible. This is the fastest way to integrate ASR without setting up infrastructure.

You will need a Typhoon API key which you can get one for free at our web playground.

Example:

See OpenAI’s API Doc for Transcription

from openai import OpenAI

client = OpenAI(
    api_key="<YOUR_API_KEY>",
    base_url="https://api.opentyphoon.ai/v1"
)

def transcribe_audio_file(audio_file_path):
    """
    Transcribe an audio file using Typhoon ASR API
    """
    try:
        with open(audio_file_path, 'rb') as audio_file:
            transcription = client.audio.transcriptions.create(
                file=audio_file,
                model="typhoon-asr-realtime"
            )
            return transcription
    except Exception as e:
        print(f"Error transcribing audio: {e}")
        return None

audio_file_path = "path/to/your/audio.wav"

transcription = transcribe_audio_file(audio_file_path)
if transcription:
    print(f"Transcription: {transcription.text}")
    print(f"Usage: {transcription.usage}")

Reference

Model ID	Size	Description	Rate Limits	Release Date
typhoon-asr-realtime	114M	Streaming ASR	100 reqs/minute	2025-09-08

🖥️ Option 2: Self-Hosting with Python Package

For developers who want to run the model locally (CPU or GPU). No API key required.

Install package

pip install typhoon-asr

Example: Local Usage

from typhoon_asr import transcribe

# Basic transcription
result = transcribe("audio.wav")
print(result['text'])

# With word timestamps
result = transcribe("audio.wav", with_timestamps=True)
for ts in result['timestamps']:
    print(f"[{ts['start']:.2f}s - {ts['end']:.2f}s] {ts['word']}")

# Specify device (CPU/GPU/auto)
result = transcribe("audio.wav", device="cuda")
print(result['text'])

API Reference (Self-Host Mode)

transcribe(
    input_file,
    model_name="scb10x/typhoon-asr-realtime",
    with_timestamps=False,
    device="auto"
)

Parameters:

• input_file (str) – Path to audio file

• model_name (str) – Hugging Face model identifier (default: scb10x/typhoon-asr-realtime)

• with_timestamps (bool) – Return word timestamps (default: False)

• device (str) – “auto”, “cpu”, “cuda”

Returns (dict):

• text – Transcribed text

• timestamps – Word timestamps (if enabled)

• processing_time – Processing duration in seconds

• audio_duration – Input audio length in seconds

Requirements

• Python ≥ 3.8

• CUDA (optional, for GPU acceleration)

Related Link

See our Github Repo for more example codes including a fine-tuning example: https://github.com/scb-10x/typhoon-asr

License

Apache Software License 2.0