Speech-to-Text API with Voice Profiling: Emotion, Accent, Age, and Pitch Detection

import requests
import base64

# pip install requests
# Auth: base64 of "<api-key>:<api-secret>" from https://platform.inworld.ai
API_KEY_B64 = "YOUR_BASE64_KEY_SECRET"

STT_ENDPOINT = "https://api.inworld.ai/stt/v1/transcribe"

# Read and base64-encode the audio file
with open("call_recording.wav", "rb") as audio_file:
    audio_b64 = base64.b64encode(audio_file.read()).decode("utf-8")

# Transcribe with Realtime STT + Voice Profile
# voiceProfileConfig.enableVoiceProfile is required to get voice profile data.
# topN (default 10) controls how many labels per category are returned.
response = requests.post(
    STT_ENDPOINT,
    headers={
        "Authorization": f"Basic {API_KEY_B64}",
        "Content-Type": "application/json"
    },
    json={
        "transcribeConfig": {
            "modelId": "inworld/inworld-stt-1",
            "language": "en",
            "audioEncoding": "AUTO_DETECT",
            "voiceProfileConfig": {
                "enableVoiceProfile": True,
                "topN": 5
            }
        },
        "audioData": {
            "content": audio_b64
        }
    },
    timeout=60
)
response.raise_for_status()

result = response.json()

# Access the transcript
transcription = result["transcription"]
print("Transcript:", transcription["transcript"])
print("Is final:", transcription["isFinal"])

# Access word-level timestamps
for word in transcription.get("wordTimestamps", []):
    print(f"  '{word['word']}' at {word['startTime']}-{word['endTime']}")

# Voice profile: emotion, accent, age, pitch, vocalStyle — each a
# {label, confidence} array sorted by descending confidence
profile = result.get("voiceProfile", {})
for category in ("emotion", "accent", "age", "pitch", "vocalStyle"):
    for entry in profile.get(category, [])[:1]:
        print(f"{category}: {entry['label']} ({entry['confidence']:.2f})")

print("Usage:", result.get("usage", {}))

import { readFile } from 'node:fs/promises';

const STT_ENDPOINT = 'https://api.inworld.ai/stt/v1/transcribe';
// Auth: base64 of "<api-key>:<api-secret>" from https://platform.inworld.ai
const API_KEY_B64 = 'YOUR_BASE64_KEY_SECRET';

async function transcribeWithProfiling(audioPath) {
  const audioBuffer = await readFile(audioPath);
  const audioB64 = audioBuffer.toString('base64');

  // voiceProfileConfig.enableVoiceProfile is required to get voice profile data.
  // topN (default 10) controls how many labels per category are returned.
  const response = await fetch(STT_ENDPOINT, {
    method: 'POST',
    headers: {
      'Authorization': `Basic ${API_KEY_B64}`,
      'Content-Type': 'application/json',
    },
    body: JSON.stringify({
      transcribeConfig: {
        modelId: 'inworld/inworld-stt-1',
        language: 'en',
        audioEncoding: 'AUTO_DETECT',
        voiceProfileConfig: {
          enableVoiceProfile: true,
          topN: 5,
        },
      },
      audioData: {
        content: audioB64,
      },
    }),
  });

  if (!response.ok) {
    throw new Error(`STT API error ${response.status}: ${await response.text()}`);
  }

  const result = await response.json();

  console.log('Transcript:', result.transcription.transcript);
  console.log('Is final:', result.transcription.isFinal);

  // Access word-level timestamps
  for (const word of result.transcription.wordTimestamps || []) {
    console.log(`  '${word.word}' at ${word.startTime}-${word.endTime}`);
  }

  // Voice profile: emotion, accent, age, pitch, vocalStyle — each a
  // {label, confidence} array sorted by descending confidence
  const profile = result.voiceProfile ?? {};
  for (const category of ['emotion', 'accent', 'age', 'pitch', 'vocalStyle']) {
    const [top] = profile[category] ?? [];
    if (top) {
      console.log(`${category}: ${top.label} (${top.confidence.toFixed(2)})`);
    }
  }

  console.log('Usage:', result.usage);

  return result;
}

transcribeWithProfiling('call_recording.wav');