← All models

Chirp 3: Transcription

Published Publisher Googlechirp_3

Model Overview

Chirp 3: Transcription is a Google generative model for multilingual automatic speech recognition (ASR) that transcribes audio into text. It is available through the Google Cloud Speech-to-Text API V2, with the model identifier chirp_3. (docs.cloud.google.com)

Capabilities and Uses

  • Supports streaming, synchronous, and batch recognition, and can be used for real-time transcription, short audio recognition, and long audio processing.
  • Supports automatic language detection, automatic punctuation, speaker diarization, and speech adaptation through word or phrase hints; language and feature availability depends on the recognition method and region. Transcription is supported for Simplified Chinese cmn-Hans-CN.
  • Can be used for transcription of meetings or recordings with multiple speakers, video captions, and content indexing. (docs.cloud.google.com)

Limitations

  • Synchronous recognition is suitable for audio shorter than 1 minute; batch recognition is generally suitable for audio from 1 minute to 1 hour, with a 20-minute limit when word-level timestamps are enabled.
  • The official documentation is inconsistent about speaker diarization: the feature table states that it applies only to BatchRecognize, while the language support section also lists Recognize. Verify support for the specific method, language, and region before use.
  • The official documentation is also inconsistent about support for word-level timestamps: one section lists them under unsupported features, while another states that they can be enabled for synchronous and batch recognition but may reduce transcription quality. The word-level confidence values returned by the API are not true confidence scores. (docs.cloud.google.com)

Release Dates

  • April 11, 2025: First released in private preview, with support for batch recognition at the time.
  • August 29, 2025: Entered public preview.
  • October 13, 2025: Generally available (GA). (docs.cloud.google.com)

Pricing

Checked on: November 18, 2026. The Google Cloud product page lists 0.016 US dollars (USD) per minute in the Speech-to-Text V2 entry that mentions Chirp 3; the service is billed by the duration of input audio successfully processed, with each request rounded up to the nearest second and multiple channels billed separately. The official page does not list a separate price for output text. The pricing details page lists tiered and dynamic batch processing prices, but does not explicitly specify the chirp_3 identifier, so whether those specific tiers apply to this model has not been verified. Other cloud services, such as storage, may incur additional charges. (cloud.google.com)

Official Sources