Gcp Speech To Text

Google Speech-to-Text V2 (with Chirp model) represents a significant accuracy improvement for challenging audio. 125+ languages, speaker diarization, word-level timestamps, and confidence scores. For agents: batch transcription for recorded files, streaming for real-time audio. V2 API with Recognizer resource management provides better accuracy than V1. Confidence is docs-derived.

Python SDK: from google.cloud.speech_v2 import SpeechClient, types. client = SpeechClient(). config = types.RecognitionConfig(auto_decoding_config=types.AutoDetectDecodingConfig(), language_codes=["en-US"], model="chirp"). audio = types.RecognitionAudio(content=audio_bytes). response = client.recognize(recognizer=f"projects/{project}/locations/global/recognizers/_", config=config, audio=audio). Streaming: client.streaming_recognize(config=streaming_config, requests=audio_generator). REST API also available.

Service account or ADC. Roles: roles/speech.client (recognize operations). ADC: gcloud auth application-default login. HTTPS enforced. Audio data handled server-side with Google privacy policies.

InvalidArgumentError for unsupported audio format. InternalError for transient service issues (retry with backoff). V2 Recognizer creation is async. Streaming: connection drops require reconnection. Speech-to-Text SLA: 99.9%. Accuracy varies by audio quality, accent, and background noise. Chirp model significantly improves accuracy over legacy models.

cloud.google.com/speech-to-text/docs covers V1 and V2 API, Chirp model, speaker diarization, custom models, and streaming. Getting started: enable Speech API, ADC configured, first recognize() in under 5 minutes. Free tier: 60 minutes/month. SDK for Python, Java, Go, Node.js. Community via Google Cloud Community.