# Google Cloud Speech-to-Text

Speech recognition API on Google Cloud that returns transcripts from audio files or live streams.

- Provider: Google Cloud (https://cloud.google.com/speech-to-text)
- Category: Transcribe audio
- Type: Hosted API
- Pricing model: Pay as you go per second of audio, with monthly volume tiers per account [1]
- Free tier: 60 minutes a month on the V1 API; the V2 price table lists none [1]
- Authentication: Google Cloud IAM: Application Default Credentials, user credentials or service accounts [8]
- Connect via: REST API (V1, V2), gRPC, client libraries, Google Cloud console [8]
- Read from the provider's pages on 2026-11-04
- Page: https://supermarket.dev/tools/transcribe-audio/google-cloud-speech-to-text

## Pricing

### Billing
- Pricing model: Pay as you go, billed per second of audio processed, with monthly tiers per account [1]
- Free tier: 60 minutes a month on the V1 API; the V2 price table lists no free tier [1]
- Free tier (new customers): $300 in Google Cloud free credits for new customers [10]
- Card required for free tier: Billing must be enabled on the Google Cloud project [10]
- Minimum spend: Not published
- Billing period: Monthly usage tiers per account; each request rounded up to 1 second [1]

### Prices
V2 `Standard` models, per account per month. $ per hour = $ per minute x 60. Dynamic batch: $0.003 x 60 = $0.18.
| Usage | $ per minute | $ per hour of audio |
|---|---|---|
| 0 to 500,000 minutes | $0.016 | $0.96 |
| 500,000 to 1,000,000 minutes | $0.01 | $0.60 |
| 1,000,000 to 2,000,000 minutes | $0.008 | $0.48 |
| 2,000,000 minutes and above | $0.004 | $0.24 |
| V2 dynamic batch (Standard models) | $0.003 | $0.18 |
| V1 with data logging, after 60 free minutes | $0.016 | $0.96 |
| V1 without data logging, after 60 free minutes | $0.024 | $1.44 |
| Medical models (V1), after 60 free minutes | $0.078 | $4.68 |
Source [1]

### Other charges
- Multiple channels: Each audio channel is billed separately [1]
- Dynamic batch: Discounted rate for audio processed at lower urgency (V2) [1]
- Volume discounts: Available for very large workloads through Google Cloud sales [1]
- Other Google Cloud services: Cloud Storage and other services used alongside are billed separately [1]
- Price tier of `chirp_2` and `chirp_3`: Not published

## Tools

### MCP server
- Official MCP server: No [19]
- Address or package: None found: no mention in the Speech-to-Text docs pages read, and no result for speech and mcp in the `googleapis` or `GoogleCloudPlatform` GitHub orgs [19]

### API
| Endpoint | What it does |
|---|---|
| `Recognize` (V2) | Synchronous recognition of short audio, inline or from Cloud Storage |
| `BatchRecognize` (V2) | Asynchronous transcription of audio files in Cloud Storage; returns an operation |
| `StreamingRecognize` (V2) | Real-time transcription of streamed audio (gRPC only) |
| `POST /v2/{parent=projects/*/locations/*}/recognizers` | Creates a recognizer, a stored recognition configuration |
| `POST /v2/{parent=projects/*/locations/*}/customClasses` | Creates a custom class for speech adaptation |
| `GET /v2/{name=projects/*}/locations` | Lists supported locations |
Source [15]

### Features
- Default model: Chosen per request in the `model` field; the docs list `chirp_3` as the latest generation [4]
- Languages: `chirp_3`: 29 locales generally available and 82 in preview (counted from its table); 125 languages across the product [17]
- Speaker labels: Yes on `chirp_3` in batch recognition; `chirp_2` does not support it. The `chirp_3` page also says `Recognize` supports it, in 15 locales [5]
- Speaker labels (other page): `chirp_2`: not supported [6]
- Word timestamps: Supported on `chirp_2`; on `chirp_3` the page lists them under unsupported features while describing them as optional for `Recognize` and `BatchRecognize` [6]
- Word timestamps (other page): `chirp_3`: listed under features it does not support [5]
- Live streaming: Yes, `StreamingRecognize` over gRPC only; a stream stays open up to 5 minutes [12]
- Max file size: Synchronous: 10 MB or 1 minute. Batch: files in Cloud Storage, each up to 8 hours [2]
- Audio input: `FLAC`, `LINEAR16`, `MULAW`, `AMR`, `AMR_WB`, `OGG_OPUS`, `SPEEX_WITH_HEADER_BYTE`, `WEBM_OPUS`; `MP3` only in `v1p1beta1` (beta) [7]

## Authentication

### Methods
| Method | Used for | How |
|---|---|---|
| Application Default Credentials | Client libraries | Libraries look for credentials in defined locations; locally set with `gcloud auth application-default login` |
| User credentials via gcloud | REST requests from the command line | `Authorization: Bearer $(gcloud auth print-access-token)` |
| Attached service account | Code running on Google Cloud | Credentials of the service account attached to the compute resource |
| Service account impersonation | Local development and testing permissions | Needs the `roles/iam.serviceAccountTokenCreator` role |
Source [8]

### Getting access
- Who creates the account: A person signs in to a Google Cloud account and selects or creates a project; agent sign-up is not stated [10]
- Where to get a key: No API key shown; credentials come from IAM (ADC, user or service account) via the `gcloud` CLI or the Google Cloud console [8]
- Other requirements: A Google Cloud project with billing enabled, and the Speech-to-Text API enabled [10]

## Setup

### Addresses
- REST API: `https://speech.googleapis.com`; regional `https://us-speech.googleapis.com` and `https://eu-speech.googleapis.com` [15]
- MCP server: None found [19]
- OpenAPI spec: Google Discovery documents (not OpenAPI): `https://speech.googleapis.com/$discovery/rest?version=v2` [15]
- Docs for agents: None: `https://cloud.google.com/llms.txt` returns HTTP 404 [19]
- Status page: `https://status.cloud.google.com/` (HTTP 200; the page lists Speech-to-Text) [19]

### Install
| What | Command |
|---|---|
| Python client library | `pip install --upgrade google-cloud-speech` |
| Node.js client library | `npm install @google-cloud/speech` |
| Go client library | `go get cloud.google.com/go/speech/apiv2` |
| CLI sign-in for local credentials | `gcloud auth application-default login` |
| MCP server | None |
Source [11]

## Limits

### Rate limits
| Limit (V2, per region, per project) | Value |
|---|---|
| Synchronous recognition requests per 60 seconds | 300 |
| Batch recognition requests per 60 seconds | 150 |
| Streaming recognition requests per 60 seconds | 1,000,000 |
| Concurrent `StreamingRecognize` sessions | 300 |
| Resource requests per 60 seconds | 100 |
| Operation requests per 60 seconds | 150 |
Source [2]

### Other limits
- Streaming: Each stream message up to 25 KB of audio; stream open up to 5 minutes [2]
- Batch: Up to 5 files per request, each up to 8 hours; Cloud Storage only [2]
- Phrases per request: 5,000 phrases and 100,000 characters in adaptation [2]
- Quota increases: Quotas are adjustable on request; system limits are fixed [2]
- V1 limits (other page): 10 MB per inline request; 900 recognition requests per 60 seconds; 480 hours of audio a day [3]

## Privacy

### Your data
- Data location: Processed globally; US or EU multi-region endpoints limit processing to those regions; single region not supported [9]
- Retention: Sync and streaming audio processed in memory; async transcripts stored about 5 days; input audio not stored [9]
- Used for training: Only for customers who opted in to data logging [9]
- Zero data retention: Not offered as a named option; default sync and streaming requests do not store customer data [9]

### Compliance
- Certifications: Not published
- HIPAA: Listed as covered by Google's Business Associate Agreement; do not opt in to data logging if a BAA covers PHI [16]

## Sources

1. Pricing: https://cloud.google.com/speech-to-text/pricing
2. Quotas and limits (V2): https://cloud.google.com/speech-to-text/v2/quotas
3. Quotas and limits (V1): https://cloud.google.com/speech-to-text/quotas
4. Transcription models (V2): https://cloud.google.com/speech-to-text/v2/docs/transcription-model
5. Chirp 3 model: https://cloud.google.com/speech-to-text/v2/docs/chirp_3-model
6. Chirp 2 model: https://cloud.google.com/speech-to-text/v2/docs/chirp_2-model
7. Audio encodings: https://cloud.google.com/speech-to-text/docs/encoding
8. Authentication: https://cloud.google.com/speech-to-text/docs/authentication
9. Data usage FAQ: https://cloud.google.com/speech-to-text/docs/data-usage-faq
10. Client libraries quickstart: https://cloud.google.com/speech-to-text/v2/docs/transcribe-client-libraries
11. Client library install: https://cloud.google.com/speech-to-text/v2/docs/libraries
12. Streaming recognition: https://cloud.google.com/speech-to-text/v2/docs/streaming-recognize
13. Batch recognition: https://cloud.google.com/speech-to-text/v2/docs/batch-recognize
14. Regional endpoints: https://cloud.google.com/speech-to-text/v2/docs/endpoints
15. REST reference: https://cloud.google.com/speech-to-text/v2/docs/reference/rest
16. HIPAA compliance on Google Cloud: https://cloud.google.com/security/compliance/hipaa
17. Google Cloud AI and ML products: https://cloud.google.com/security/compliance/offerings
18. Google Cloud status: https://status.cloud.google.com/
19. Checked by supermarket.dev