SaluteSpeech (Sberbank) Integration for Speech Recognition Salute

Speech is Sberbank's speech technology with a focus on the Russian language. Russian infrastructure, GOST compliance, and the ability to deploy on-premise. Particularly suitable for the banking, financial, and government sectors. ### Technical Specifications - WER in spoken Russian: 10–14% - Streaming Recognition Latency: 200–400 ms - Support for 8 kHz and 16 kHz audio - Diarization for up to 10 speakers ### REST API Integration```python import requests import base64

Получение токена

def get_token(): response = requests.post( "https://ngw.devices.sberbank.ru:9443/api/v2/oauth", headers={ "Authorization": f"Basic {base64.b64encode(f'{CLIENT_ID}:{CLIENT_SECRET}'.encode()).decode()}", "RqUID": "unique-request-id", "Content-Type": "application/x-www-form-urlencoded" }, data={"scope": "SALUTE_SPEECH_PERS"} ) return response.json()["access_token"]

Транскрипция

def transcribe(audio_bytes: bytes, token: str): response = requests.post( "https://smartspeech.sber.ru/rest/v1/speech:recognize", headers={ "Authorization": f"Bearer {token}", "Content-Type": "audio/x-pcm;bit=16;rate=16000" }, data=audio_bytes ) return response.json()["result"][0]["normalized_text"]