This document contains the current API restrictions and usage limits forCloud Speech-to-Text. This page will be updated to reflect any changes to theserestrictions and usage limits. We reserve the right to change these limits.

You canrequest a quotaincrease ifnecessary. See the Google Cloudquota page for moreinformation on viewing and managing your quota.

After submitting your request, Google might contact you for more information,and inform you whether your request is approved or denied.

Note: The time limits listed below refer to the length of the audio file,regardless of whether it ismultichannel audio. Beaware, however, that each channel is billed individually. This can lead tosituations where you are billed for more time than than the duration of theaudio file. See thepricing page for more information.

Content limits

Synchronous requests

Synchronous recognition requests (using theRecognize method) accept audio data either inline in thecontent field of the request or as aCloud Storage URI in theuri field of the request. Audio sent to a synchronousrequest is limited to 10 MB or 1 minute of audio duration (whichever is reachedfirst). For more information on synchronous recognition, see thesynchronousrecognition overview.

Streaming requests

Streaming recognition requests (using theStreamingRecognize method) only accept inline audio in theaudio field of the request. Each requestin the stream is limited to 25 KB of audio. A stream can remain open for up to 5minutes, and the audio must be sent at a rate that approximates real time. Ifyou need to stream content for longer than 5 minutes, see theendless streamingtutorial. For more information on streamingrecognition, see thestreaming recognition overview.

Batch requests

Batch recognition requests (using theBatchRecognize method) only accept audio as a Cloud StorageURI in theuri fieldof the request. EachBatchRecognizeRequestcan contain up to 15files to transcribe.Each file can be up to 8 hours in duration. For more information on asynchronousrecognition, see thebatch recognition overview.

Multiple language recognition

Multiple language recognition is only available in the global, US, and EUCloud Speech-to-Text endpoints.

Adaptation

Within any request, you may also supplyPhraseSet andCustomClass resources. The following limits apply to theseresources:

Speech Adaptation Limit	Value
Maximum allowable phrase boost value	20
Phrases in a PhraseSet	1,200
Phrases per request	5,000
Characters per phrase	100
Total characters per request	100,000
Maximum number of items in a CustomClass	500
Maximum characters per CustomClass item	500
Maximum number of PhraseSets per SpeechAdaptation	20
Maximum number of CustomClasses per SpeechAdaptation	20

Resource limits

The current API resource limits for Cloud Speech-to-Text are as follows (and aresubject to change):

Type of Limit	Usage Limit
Number of recognizers (per region)	5,000
Number of custom classes (per region)	5,000
Number of phrase sets (per region)	5,000

Request limits

The current API usage limits for Cloud Speech-to-Text are as follows (and are subjectto change):

Type of Limit	Usage Limit
Resource requests per 60 seconds (per region)	100
Operation requests per 60 seconds (per region)	150
Synchronous recognition requests per 60 seconds (per region)	300
Streaming recognition requests per 60 seconds (per region) *	1,000,000
Concurrent StreamingRecognize sessions (per region) *	300
Batch recognition requests per 60 seconds (per region)	150

* Streaming recognition has a quota limit of 300 concurrent sessions per 5minutes and a limit of 3,000 requests per minute, which applies to allconcurrent sessions together. The initial configuration request for a sessiondoes not count against the request quota.

These limits apply to each Cloud Speech-to-Text developer project, and are sharedacross all applications and IP addresses using a given a developer project.