Потоковая транскрибация

wss://api.palatine.ru/api/v1/stream/asr?token=<YOUR_TOKEN>&sample_rate=48000&codec=linear16

GET wss://api.palatine.ru/api/v1/stream/asr?sample_rate=48000&codec=linear16
Authorization: Bearer <YOUR_TOKEN>

{
  "type": "config",
  "model": "palatine_stream",
  "sample_rate": 48000,
  "channels": 1,
  "format": "linear16",
  "language": "ru",
  "recommended_codec": "linear16",
  "supported_codecs": ["linear16", "linear32", "flac", "alaw", "mulaw", "amr-nb", "amr-wb", "g729", "opus", "ogg-opus", "speex", "mp3", "mp2", "aac", "webm", "mp4", "ogg", "wav", "m4a"],
  "supported_sample_rates": [8000, 16000, 22050, 32000, 44100, 48000]
}

import asyncio
import websockets
import wave

API_URL = "wss://api.palatine.ru/api/v1/stream/asr"
TOKEN = "<YOUR_TOKEN>"
SAMPLE_RATE = 48000  # Частота дискретизации аудиофайла

async def stream_audio():
    url = f"{API_URL}?token={TOKEN}&sample_rate={SAMPLE_RATE}&codec=linear16"

    async with websockets.connect(url) as ws:
        # Получаем конфигурацию
        config = await ws.recv()
        print(f"Конфигурация: {config}")

        # Читаем и отправляем аудиофайл
        with wave.open("audio.wav", "rb") as wf:
            # 150ms при текущем sample_rate (sample_rate * 0.15 * 2 bytes)
            chunk_size = int(SAMPLE_RATE * 0.15 * 2)
            while True:
                data = wf.readframes(chunk_size // 2)
                if not data:
                    break
                await ws.send(data)

                # Получаем промежуточные результаты
                try:
                    result = await asyncio.wait_for(ws.recv(), timeout=0.1)
                    print(result)
                except asyncio.TimeoutError:
                    pass

        # Сигнализируем об окончании потока
        await ws.send(b'')

        # Получаем оставшиеся результаты
        async for message in ws:
            print(message)

asyncio.run(stream_audio())

import asyncio
import websockets

API_URL = "wss://api.palatine.ru/api/v1/stream/asr"
TOKEN = "<YOUR_TOKEN>"
SAMPLE_RATE = 16000
CHUNK_SIZE = int(SAMPLE_RATE * 0.3 * 2)  # 300 мс = 9600 байт

async def stream_audio():
    url = f"{API_URL}?sample_rate={SAMPLE_RATE}&codec=linear16&channels=1"

    async with websockets.connect(
        url, additional_headers={"Authorization": f"Bearer {TOKEN}"}
    ) as ws:
        print(await ws.recv())  # {"type": "config", ...}

        async def send_audio():
            # Сырой PCM16 LE без контейнера. Для .wav используйте
            # wave.open() (как в примере выше) либо укажите codec=wav.
            with open("audio.raw", "rb") as f:
                while chunk := f.read(CHUNK_SIZE):
                    await ws.send(chunk)
                    await asyncio.sleep(0.3)  # темп реального времени
            await ws.send(b"")  # конец потока

        async def receive():
            async for message in ws:
                print(message)

        await asyncio.gather(send_audio(), receive())

asyncio.run(stream_audio())

{
  "type": "partial",
  "chunk_text": " мир",
  "full_text": "привет мир",
  "words": [{"word": "привет", "start": 0.42, "end": 0.9}],
  "phrases": []
}

{
  "type": "partial",
  "chunk_text": "",
  "full_text": "привет мир",
  "words": [{"word": "мир", "start": 0.95, "end": 1.31}],
  "phrases": [{"text": "привет мир", "start": 0.42, "end": 1.31}]
}

{
  "type": "final",
  "status": "completed",
  "total_duration": 15.5
}

wss://api.palatine.ru/api/v1/stream/asr?token=<YOUR_TOKEN>&sample_rate=48000&codec=linear16&formatting=true

{
  "type": "partial",
  "chunk_text": " тридцать",
  "full_text": "встреча двадцать пятого мая в четырнадцать тридцать",
  "words": [{"word": "четырнадцать", "start": 1.98, "end": 2.55}],
  "phrases": [],
  "formatted": {
    "text": "Встреча 25 мая в четырнадцать тридцать",
    "committed": ""
  }
}

{
  "formatted": {
    "text": "Встреча 25 мая в 14:30.",
    "committed": "Встреча 25 мая в 14:30."
  }
}

{
  "type": "error",
  "message": "ML service unavailable. Please try again later.",
  "code": "ML_SERVICE_ERROR"
}

{
  "type": "error",
  "message": "Unauthorized: Invalid or missing token",
  "code": "AUTH_ERROR"
}

import asyncio
import websockets
import pyaudio

API_URL = "wss://api.palatine.ru/api/v1/stream/asr"
TOKEN = "<YOUR_TOKEN>"

SAMPLE_RATE = 48000
CHANNELS = 1
CHUNK_SIZE = int(SAMPLE_RATE * 0.15 * 2)  # 150ms

async def transcribe_microphone():
    url = f"{API_URL}?token={TOKEN}&sample_rate={SAMPLE_RATE}&codec=linear16"

    # Инициализация PyAudio
    p = pyaudio.PyAudio()
    stream = p.open(
        format=pyaudio.paInt16,
        channels=CHANNELS,
        rate=SAMPLE_RATE,
        input=True,
        frames_per_buffer=CHUNK_SIZE // 2
    )

    print("Начинаю запись... Нажмите Ctrl+C для остановки.")

    try:
        async with websockets.connect(url) as ws:
            # Получаем конфигурацию
            config = await ws.recv()
            print(f"Подключено: {config}")

            async def send_audio():
                while True:
                    data = stream.read(CHUNK_SIZE // 2, exception_on_overflow=False)
                    await ws.send(data)
                    await asyncio.sleep(0.01)

            async def receive_results():
                async for message in ws:
                    print(message)

            await asyncio.gather(
                send_audio(),
                receive_results()
            )

    except KeyboardInterrupt:
        print("\nОстановка...")
    finally:
        stream.stop_stream()
        stream.close()
        p.terminate()

asyncio.run(transcribe_microphone())

class PalatineASR {
  constructor(token, sampleRate = 48000) {
    this.token = token;
    this.sampleRate = sampleRate;
    this.ws = null;
    this.audioContext = null;
    this.mediaStream = null;
    this.workletNode = null;
    this.onResult = null;
    this.onError = null;
  }

  async start() {
    // Запрашиваем доступ к микрофону
    this.mediaStream = await navigator.mediaDevices.getUserMedia({
      audio: {
        sampleRate: this.sampleRate,
        channelCount: 1,
        echoCancellation: true,
        noiseSuppression: true
      }
    });

    // Создаём AudioContext
    this.audioContext = new AudioContext({ sampleRate: this.sampleRate });

    // Подключаемся к WebSocket
    const url = `wss://api.palatine.ru/api/v1/stream/asr?token=${this.token}&sample_rate=${this.sampleRate}&codec=linear16`;
    this.ws = new WebSocket(url);

    this.ws.onmessage = (event) => {
      const data = JSON.parse(event.data);
      if (data.type === 'partial' && this.onResult) {
        this.onResult(data);
      } else if (data.type === 'error' && this.onError) {
        this.onError(data);
      }
    };

    this.ws.onerror = (error) => {
      if (this.onError) this.onError({ message: 'WebSocket error', error });
    };

    // Ждём открытия соединения
    await new Promise((resolve, reject) => {
      this.ws.onopen = resolve;
      this.ws.onerror = reject;
    });

    // Загружаем AudioWorklet для обработки аудио
    await this.audioContext.audioWorklet.addModule(this.createWorkletURL());

    // Создаём узлы обработки
    const source = this.audioContext.createMediaStreamSource(this.mediaStream);
    this.workletNode = new AudioWorkletNode(this.audioContext, 'pcm-processor');

    this.workletNode.port.onmessage = (event) => {
      if (event.data?.type === 'flushed') {
        if (this.ws?.readyState === WebSocket.OPEN) this.ws.send(new ArrayBuffer(0));
        this.workletNode.disconnect();
        this.audioContext.close();
        return;
      }
      if (this.ws?.readyState === WebSocket.OPEN) this.ws.send(event.data);
    };

    source.connect(this.workletNode);
    this.workletNode.connect(this.audioContext.destination);
  }

  createWorkletURL() {
    const bufferSize = Math.floor(this.sampleRate * 0.15); // 150ms буфер
    const workletCode = `
      class PCMProcessor extends AudioWorkletProcessor {
        constructor() {
          super();
          this.buffer = [];
            this.stopped = false;
            this.port.onmessage = ({ data }) => {
              if (data !== 'flush' || this.stopped) return;
              this.stopped = true;
              if (this.buffer.length) {
                this.port.postMessage(new Int16Array(this.buffer.splice(0)).buffer);
              }
              this.port.postMessage({ type: 'flushed' });
            };
          this.bufferSize = ${bufferSize};
        }

        process(inputs) {
          if (this.stopped) return false;
          const input = inputs[0];
          if (input.length > 0) {
            const samples = input[0];
            for (let i = 0; i < samples.length; i++) {
              // Конвертируем float32 [-1, 1] в int16
              const s = Math.max(-1, Math.min(1, samples[i]));
              this.buffer.push(s < 0 ? s * 0x8000 : s * 0x7FFF);
            }

            while (this.buffer.length >= this.bufferSize) {
              const chunk = this.buffer.splice(0, this.bufferSize);
              const int16Array = new Int16Array(chunk);
              this.port.postMessage(int16Array.buffer);
            }
          }
          return true;
        }
      }
      registerProcessor('pcm-processor', PCMProcessor);
    `;
    const blob = new Blob([workletCode], { type: 'application/javascript' });
    return URL.createObjectURL(blob);
  }

  stop() {
    this.mediaStream?.getTracks().forEach(track => track.stop());
    // Сначала досылаем остаток PCM; EOS уйдёт после подтверждения flushed.
    this.workletNode?.port.postMessage('flush');
    // WebSocket остаётся открыт для последних partial и final.
  }
}

// Использование
const asr = new PalatineASR('<YOUR_TOKEN>');

asr.onResult = (data) => {
  console.log('Распознано:', data.full_text);
  document.getElementById('transcript').textContent = data.full_text;
};

asr.onError = (error) => {
  console.error('Ошибка:', error);
};

// Запуск по кнопке
document.getElementById('startBtn').onclick = () => asr.start();
document.getElementById('stopBtn').onclick = () => asr.stop();

<button id="startBtn">Начать запись</button>
<button id="stopBtn">Остановить</button>
<div id="transcript"></div>

import { useState, useRef, useCallback } from 'react';

interface TranscriptResult {
  type: 'partial' | 'final' | 'error';
  chunk_text?: string;
  full_text?: string;
  words?: { word: string; start: number; end: number }[];
  phrases?: { text: string; start: number; end: number }[];
  message?: string;
}

const SAMPLE_RATE = 48000;

export function useStreamingASR(token: string) {
  const [isRecording, setIsRecording] = useState(false);
  const [transcript, setTranscript] = useState('');
  const [error, setError] = useState<string | null>(null);

  const wsRef = useRef<WebSocket | null>(null);
  const audioContextRef = useRef<AudioContext | null>(null);
  const workletRef = useRef<AudioWorkletNode | null>(null);
  const streamRef = useRef<MediaStream | null>(null);

  const start = useCallback(async () => {
    try {
      setError(null);
      setTranscript('');

      // Запрашиваем микрофон
      const stream = await navigator.mediaDevices.getUserMedia({
        audio: { sampleRate: SAMPLE_RATE, channelCount: 1 }
      });
      streamRef.current = stream;

      // Создаём AudioContext
      const audioContext = new AudioContext({ sampleRate: SAMPLE_RATE });
      audioContextRef.current = audioContext;

      // WebSocket подключение
      const ws = new WebSocket(
        `wss://api.palatine.ru/api/v1/stream/asr?token=${token}&sample_rate=${SAMPLE_RATE}&codec=linear16`
      );
      wsRef.current = ws;
      ws.onclose = () => setIsRecording(false);

      ws.onmessage = (event) => {
        const data: TranscriptResult = JSON.parse(event.data);
        if (data.type === 'partial' && data.full_text) {
          setTranscript(data.full_text);
        } else if (data.type === 'error') {
          setError(data.message || 'Неизвестная ошибка');
        }
      };

      await new Promise<void>((resolve, reject) => {
        ws.onopen = () => resolve();
        ws.onerror = () => reject(new Error('Ошибка подключения'));
      });

      // AudioWorklet для захвата PCM
      const bufferSize = Math.floor(SAMPLE_RATE * 0.15); // 150ms
      const workletCode = `
        class PCMProcessor extends AudioWorkletProcessor {
          constructor() {
            super();
            this.buffer = [];
            this.stopped = false;
            this.port.onmessage = ({ data }) => {
              if (data !== 'flush' || this.stopped) return;
              this.stopped = true;
              if (this.buffer.length) {
                this.port.postMessage(new Int16Array(this.buffer.splice(0)).buffer);
              }
              this.port.postMessage({ type: 'flushed' });
            };
          }
          process(inputs) {
            if (this.stopped) return false;
            const input = inputs[0];
            if (input.length > 0) {
              for (const sample of input[0]) {
                const s = Math.max(-1, Math.min(1, sample));
                this.buffer.push(s < 0 ? s * 0x8000 : s * 0x7FFF);
              }
              if (this.buffer.length >= ${bufferSize}) {
                const chunk = this.buffer.splice(0, ${bufferSize});
                this.port.postMessage(new Int16Array(chunk).buffer);
              }
            }
            return true;
          }
        }
        registerProcessor('pcm-processor', PCMProcessor);
      `;
      const blob = new Blob([workletCode], { type: 'application/javascript' });
      await audioContext.audioWorklet.addModule(URL.createObjectURL(blob));

      const source = audioContext.createMediaStreamSource(stream);
      const worklet = new AudioWorkletNode(audioContext, 'pcm-processor');
      workletRef.current = worklet;

      worklet.port.onmessage = (e) => {
        if (e.data?.type === 'flushed') {
          if (ws.readyState === WebSocket.OPEN) ws.send(new ArrayBuffer(0));
          worklet.disconnect();
          audioContext.close();
          return;
        }
        if (ws.readyState === WebSocket.OPEN) ws.send(e.data);
      };

      source.connect(worklet);
      setIsRecording(true);

    } catch (err) {
      setError(err instanceof Error ? err.message : 'Ошибка запуска');
    }
  }, [token]);

  const stop = useCallback(() => {
    streamRef.current?.getTracks().forEach(t => t.stop());
    workletRef.current?.port.postMessage('flush');
    // После flushed отправится EOS; состояние сбросится при закрытии сервером.
  }, []);

  return { isRecording, transcript, error, start, stop };
}

// Пример использования
export function TranscriptionWidget({ token }: { token: string }) {
  const { isRecording, transcript, error, start, stop } = useStreamingASR(token);

  return (
    <div>
      <button onClick={isRecording ? stop : start}>
        {isRecording ? 'Остановить' : 'Начать запись'}
      </button>
      {error && <p style={{ color: 'red' }}>{error}</p>}
      <p>{transcript || 'Говорите...'}</p>
    </div>
  );
}

<template>
  <div class="transcription-widget">
    <button @click="toggle" :disabled="connecting">
      {{ isRecording ? 'Остановить' : 'Начать запись' }}
    </button>
    <p v-if="error" class="error">{{ error }}</p>
    <p class="transcript">{{ transcript || 'Говорите...' }}</p>
  </div>
</template>

<script setup lang="ts">
import { ref, onUnmounted } from 'vue';

const props = defineProps<{ token: string }>();

const SAMPLE_RATE = 48000;
const BUFFER_SIZE = Math.floor(SAMPLE_RATE * 0.15); // 150ms

const isRecording = ref(false);
const connecting = ref(false);
const transcript = ref('');
const error = ref<string | null>(null);

let ws: WebSocket | null = null;
let audioContext: AudioContext | null = null;
let mediaStream: MediaStream | null = null;
let workletNode: AudioWorkletNode | null = null;

async function start() {
  try {
    connecting.value = true;
    error.value = null;
    transcript.value = '';

    mediaStream = await navigator.mediaDevices.getUserMedia({
      audio: { sampleRate: SAMPLE_RATE, channelCount: 1 }
    });

    audioContext = new AudioContext({ sampleRate: SAMPLE_RATE });

    ws = new WebSocket(
      `wss://api.palatine.ru/api/v1/stream/asr?token=${props.token}&sample_rate=${SAMPLE_RATE}&codec=linear16`
    );

    ws.onclose = () => { isRecording.value = false; };
    ws.onmessage = (event) => {
      const data = JSON.parse(event.data);
      if (data.type === 'partial') {
        transcript.value = data.full_text;
      } else if (data.type === 'error') {
        error.value = data.message;
      }
    };

    await new Promise<void>((resolve, reject) => {
      ws!.onopen = () => resolve();
      ws!.onerror = () => reject(new Error('Ошибка подключения'));
    });

    const workletCode = `
      class PCMProcessor extends AudioWorkletProcessor {
        constructor() {
          super();
          this.buffer = [];
            this.stopped = false;
            this.port.onmessage = ({ data }) => {
              if (data !== 'flush' || this.stopped) return;
              this.stopped = true;
              if (this.buffer.length) {
                this.port.postMessage(new Int16Array(this.buffer.splice(0)).buffer);
              }
              this.port.postMessage({ type: 'flushed' });
            };
        }
        process(inputs) {
          if (this.stopped) return false;
          const input = inputs[0];
          if (input.length > 0) {
            for (const s of input[0]) {
              const v = Math.max(-1, Math.min(1, s));
              this.buffer.push(v < 0 ? v * 0x8000 : v * 0x7FFF);
            }
            if (this.buffer.length >= ${BUFFER_SIZE}) {
              this.port.postMessage(new Int16Array(this.buffer.splice(0, ${BUFFER_SIZE})).buffer);
            }
          }
          return true;
        }
      }
      registerProcessor('pcm-processor', PCMProcessor);
    `;
    const blob = new Blob([workletCode], { type: 'application/javascript' });
    await audioContext.audioWorklet.addModule(URL.createObjectURL(blob));

    const source = audioContext.createMediaStreamSource(mediaStream);
    const worklet = new AudioWorkletNode(audioContext, 'pcm-processor');
    workletNode = worklet;

    worklet.port.onmessage = (e) => {
      if (e.data?.type === 'flushed') {
        if (ws?.readyState === WebSocket.OPEN) ws.send(new ArrayBuffer(0));
        worklet.disconnect();
        audioContext?.close();
        return;
      }
      if (ws?.readyState === WebSocket.OPEN) ws.send(e.data);
    };

    source.connect(worklet);
    isRecording.value = true;

  } catch (err) {
    error.value = err instanceof Error ? err.message : 'Ошибка';
  } finally {
    connecting.value = false;
  }
}

function stop() {
  mediaStream?.getTracks().forEach(t => t.stop());
  workletNode?.port.postMessage('flush');
  // После flushed отправится EOS; состояние сбросится при закрытии сервером.
}

function toggle() {
  isRecording.value ? stop() : start();
}

onUnmounted(() => stop());
</script>

<style scoped>
.error { color: red; }
.transcript { min-height: 100px; padding: 1rem; background: #f5f5f5; }
</style>