音声認識API

curl --location 'https://api.shisa.ai/asr/srt/audio_llm' \
  --header 'Authorization: Bearer shsk:YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "shisa-ai/shisa-asr-v0.1b",
    "temperature": 0.0,
    "top_p": 0.85,
    "frequency_penalty": 0.5,
    "repetition_penalty": 1.05,
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Transcribe the Japanese audio clip into text. Here are hotwords that may appear: Shisa AI, API, 音声認識"
          },
          {
            "type": "audio_url",
            "audio_url": {
              "url": "data:audio/ogg;base64,'$(base64 -w0 audio.ogg)'"
            }
          }
        ]
      }
    ]
  }'

{
  "language": "auto",
  "text": "祖母はおおむね機嫌よくさえころころがしている。"
}

{
  "language": "auto",
  "text": "祖母はおおむね機嫌よくさえころころがしている。"
}

curl --location 'https://api.shisa.ai/asr/srt/audio_llm' \
  --header 'Authorization: Bearer shsk:YOUR_API_KEY' \
  --header 'Content-Type: application/json' \
  --data '{
    "model": "shisa-ai/shisa-asr-v0.1b",
    "temperature": 0.0,
    "top_p": 0.85,
    "frequency_penalty": 0.5,
    "repetition_penalty": 1.05,
    "messages": [
      {
        "role": "user",
        "content": [
          {
            "type": "text",
            "text": "Transcribe the Japanese audio clip into text. Here are hotwords that may appear: Shisa AI, API, 音声認識"
          },
          {
            "type": "audio_url",
            "audio_url": {
              "url": "data:audio/ogg;base64,'$(base64 -w0 audio.ogg)'"
            }
          }
        ]
      }
    ]
  }'

import base64
import json
import requests

def transcribe_audio(audio_path, language="Japanese", hotwords=None):
    """Transcribe audio file using Shisa AI ASR API"""

    # Read and encode audio file
    with open(audio_path, 'rb') as audio_file:
        audio_data = base64.b64encode(audio_file.read()).decode('utf-8')

    # Determine audio format from file extension
    audio_format = audio_path.split('.')[-1]

    # Build hotwords text
    hotwords_text = ""
    if hotwords:
        hotwords_text = f"Here are hotwords that may appear: {', '.join(hotwords)}"

    # Prepare request
    url = "https://api.shisa.ai/asr/srt/audio_llm"
    headers = {
        "Authorization": "Bearer shsk:YOUR_API_KEY",
        "Content-Type": "application/json"
    }

    payload = {
        "model": "shisa-ai/shisa-asr-v0.1b",
        "temperature": 0.0,
        "top_p": 0.85,
        "frequency_penalty": 0.5,
        "repetition_penalty": 1.05,
        "messages": [
            {
                "role": "user",
                "content": [
                    {
                        "type": "text",
                        "text": f"Transcribe the {language} audio clip into text. {hotwords_text}"
                    },
                    {
                        "type": "audio_url",
                        "audio_url": {
                            "url": f"data:audio/{audio_format};base64,{audio_data}"
                        }
                    }
                ]
            }
        ]
    }

    # Make request
    response = requests.post(url, headers=headers, json=payload)
    response.raise_for_status()

    return response.json()

# Example usage
result = transcribe_audio(
    "meeting_recording.ogg",
    language="Japanese",
    hotwords=["Shisa AI", "API", "音声認識"]
)

print(result)

async function transcribeAudio(audioFile, language = 'Japanese', hotwords = []) {
  // Read file and convert to base64
  const fileBuffer = await audioFile.arrayBuffer();
  const base64Audio = btoa(
    new Uint8Array(fileBuffer).reduce(
      (data, byte) => data + String.fromCharCode(byte),
      ''
    )
  );

  // Get audio format from file type
  const audioFormat = audioFile.type.split('/')[1];

  // Build hotwords text
  const hotwordsText = hotwords.length > 0
    ? `Here are hotwords that may appear: ${hotwords.join(', ')}`
    : '';

  // Prepare request
  const response = await fetch('https://api.shisa.ai/asr/srt/audio_llm', {
    method: 'POST',
    headers: {
      'Authorization': 'Bearer shsk:YOUR_API_KEY',
      'Content-Type': 'application/json'
    },
    body: JSON.stringify({
      model: 'shisa-ai/shisa-asr-v0.1b',
      temperature: 0.0,
      top_p: 0.85,
      frequency_penalty: 0.5,
      repetition_penalty: 1.05,
      messages: [
        {
          role: 'user',
          content: [
            {
              type: 'text',
              text: `Transcribe the ${language} audio clip into text. ${hotwordsText}`
            },
            {
              type: 'audio_url',
              audio_url: {
                url: `data:audio/${audioFormat};base64,${base64Audio}`
              }
            }
          ]
        }
      ]
    })
  });

  if (!response.ok) {
    throw new Error(`API request failed: ${response.status}`);
  }

  return await response.json();
}

// Example usage with file input
document.querySelector('#audioInput').addEventListener('change', async (e) => {
  const file = e.target.files[0];
  if (file) {
    const result = await transcribeAudio(
      file,
      'Japanese',
      ['Shisa AI', 'API', '音声認識']
    );
    console.log('Transcription:', result);
  }
});