BestOnline.ToolsBestOnline.Tools
Voice generation

Google Cloud Text-to-Speech

Google Cloud Text-to-Speech: Chirp voices and Gemini narration

Google Cloud Text-to-Speech creates multilingual narration and spoken app responses with Chirp 3 HD, prompt-directed Gemini speech and usage-based API pricing.

Multilingual speech APIPrompted deliveryStreaming audio

Free plan availableAccount required

Product previewView original
Google’s example architecture connects Text-to-Speech to an application, audio storage and delivery. The illustrated Cloud Run, Storage and CDN services have separate costs.
Google’s example architecture connects Text-to-Speech to an application, audio storage and delivery. The illustrated Cloud Run, Storage and CDN services have separate costs.Google Cloud
Access
Free and paid plans
Usage pricing
USD 4 / 1,000,000 charactersStandard / WaveNet
Available on
Web, API
OVERVIEW

Cloud speech for narration and interactive applications

Google Cloud Text-to-Speech fits developers adding narration, spoken notifications and interactive responses to an application. Its REST and gRPC APIs sit alongside Google Cloud billing and authentication. Chirp 3 HD provides multilingual stock voices; Gemini TTS accepts natural-language instructions for tone, pace, accent and expression. Standard, WaveNet, Neural2 and Studio offer other voice and control options. The service requires a billing-enabled project, including when your usage stays inside a free allowance.

Gemini 2.5 Flash and Pro support single-speaker narration and multi-speaker dialogue with a voice assigned to each speaker. The Preview 2.5 Flash-Lite model supports a single speaker, while Preview 3.1 Flash supports dialogue. Cloud Text-to-Speech and Vertex AI have different response formats and streaming behavior. Gemini 3.8 Flash and Flash-Lite speech, including voice design and replication, use the separate Gemini Enterprise API in Preview; they are not Cloud Text-to-Speech API models.

Chirp 3 HD offers Preview pace, pause and pronunciation controls. Speaking rate spans 0.25 to 2 times normal speed. Pause tags belong in the markup input, and their duration varies with context; the model can occasionally ignore them. IPA and X-SAMPA pronunciation overrides help with individual words, subject to locale availability. Chirp SSML is also in Preview and applies to synchronous requests, not streaming. Studio has its own SSML restrictions, so controls are not interchangeable across voice families.

Streaming can start returning audio while an application is still sending text. Cloud TTS supports bidirectional streaming for its supported voices, whereas Gemini through Vertex AI takes one request and streams the response. Ordinary Cloud TTS requests have a 5,000-byte content limit; the Gemini Cloud TTS route separately caps text and delivery prompts at 4,000 bytes each and 8,000 bytes combined. Gemini audio can be truncated at approximately 655 seconds. A separate Preview long-audio operation accepts up to 1 million input bytes and writes to Cloud Storage; it does not remove Gemini model limits.

Character-priced voices and Gemini speech have different costs. Standard and WaveNet list at USD 4 per million characters after the shared 4-million monthly allowance; Chirp 3 HD lists at USD 30 after its 1-million allowance. Gemini charges for text input tokens and audio output tokens separately. Instant Custom Voice requires sales allowlisting, prescribed recorded consent and a short reference clip, with no free character allowance. Cloud storage and other infrastructure are billed separately. Google’s no-logging statement for Cloud TTS synthesis text and audio should not be treated as a blanket policy for custom voice recordings or other Gemini services.

Best for

  • Developers adding multilingual speech to applications
  • Teams producing narration and spoken customer responses
  • Products needing prompted delivery or streaming dialogue

Limitations

Billing is required and usage above free allowances is charged automatically. Streaming formats, pronunciation controls, language coverage and request limits vary by model and API. Chirp SSML and advanced controls are in Preview. Custom Voice requires sales approval and recorded consent. Gemini 3.8 speech uses the separate Gemini Enterprise API.

WHAT YOU CAN DO

Google Cloud speech-generation features

Multilingual speech through a cloud API

Generate narration and spoken application responses with Chirp 3 HD, Gemini or legacy voice families. Available languages, regional variants and launch stages depend on the model.

Prompt-directed Gemini narration and dialogue

Guide tone, pace, accent and expression through a delivery prompt. Gemini 2.5 Flash and Pro support single- and multi-speaker speech; Preview Flash-Lite is single-speaker.

Chirp pronunciation and pacing controls

Preview controls support 0.25–2× speaking rate, markup pause tags and IPA/X-SAMPA word pronunciations. Pause lengths are contextual and some controls have locale exclusions; Preview SSML works only for synchronous requests.

Bidirectional streaming and separate output formats

Cloud TTS can receive text chunks while returning audio. Gemini unary output includes MP3, LINEAR16, PCM, Ogg Opus and telephony encodings; Gemini streaming excludes MP3. Vertex AI returns raw 16-bit 24 kHz PCM without a WAV header.

Long-audio synthesis to Cloud Storage

A separate Preview asynchronous operation accepts up to 1 million input bytes and writes audio to a Cloud Storage bucket. Its supported voice and operation requirements differ from a normal Gemini request.

Consent-based custom voices

Chirp 3 Instant Custom Voice requires sales allowlisting, a prescribed consent recording and a reference clip, each up to 10 seconds. Both recordings must be made in the same environment; the resulting voice cloning key is stored by the client.

Usage pricing with model-specific free allowances

Character-priced models have their own allowances and rates; Gemini bills input and audio output tokens separately. Billing must be enabled, and Instant Custom Voice and Gemini have no listed free usage allowance.

PLANS & PRICING

Google Cloud Text-to-Speech pricing

Usage-based pricing, not a monthly subscription. Standard/WaveNet lists at USD 4 per million characters after its shared 4-million monthly allowance; Chirp 3 HD is 30 after its 1-million allowance. Neural2 is 16 and Studio 160 after their listed allowances. Instant Custom Voice is 60 with no free allowance. Gemini bills text input and audio output tokens separately; other Cloud resources cost extra.

Example plans and usage rates; additional charges may apply.

Standard / WaveNet

USD 4 / 1,000,000 characters

Plan details

Per million characters beyond the listed 4-million monthly allowance; same pricing SKU, not two additive free quotas.

Neural2

USD 16 / 1,000,000 characters

Plan details

Per million characters beyond the listed 1-million monthly allowance.

Chirp 3 HD

USD 30 / 1,000,000 characters

Plan details

Per million characters beyond the listed 1-million monthly allowance.

Pricing updated 2026-10-09. Prices may change.View current pricing →
Technical specifications

Voice generation

FeatureGoogle Cloud Text-to-Speech
Speech languages
Multilingual voices and regional locales across Chirp, Gemini and legacy families. Language coverage and Preview availability depend on the selected model; custom-voice locale support is separate.Model-specific language and regional availability
Voice and pronunciation controls
Chirp 3 HD SSML and pace, pause and pronunciation controls are in Preview. SSML applies to synchronous requests only. Pace spans 0.25–2×; pause tags use markup and are not fixed-duration, with locale exceptions for pause and pronunciation controls.Advanced controls Preview; availability varies by locale
Voice cloning
Chirp 3 Instant Custom Voice requires sales allowlisting, a prescribed recorded-consent statement and a reference clip, each up to 10 seconds. Supported languages and regions are limited.Sales approval required; recorded speaker consent
Audio exports and streaming
Cloud TTS Gemini unary requests support LINEAR16, PCM, MP3, Ogg Opus and A-law/μ-law. Streaming supports PCM, Ogg Opus and telephony encodings. Vertex AI returns raw 16-bit 24 kHz PCM.Encoding and streaming support depend on API and model
Commercial-use terms
Google Cloud AI terms treat generative output as Customer Data and do not assert ownership of newly created output IP. Service restrictions and rights to supplied text or voice recordings still apply.Cloud agreement and applicable use restrictions
API access
SupportedBilling-enabled Cloud project and authentication required
MORE AI TOOLS

Google Cloud Text-to-Speech alternatives

Other services offer AWS integration, managed streaming voices or creator-focused speech tools.

Amazon Polly

Amazon Polly

Four AWS speech engines with SSML, lexicons and per-character billing.

Cartesia

Cartesia

Sonic streaming speech and custom voices with shared speech credits.

ElevenLabs

ElevenLabs

Hosted speech and cloning with a browser studio and other audio production tools.

COMMON QUESTIONS

Google Cloud Text-to-Speech FAQs

Is Google Cloud Text-to-Speech free?

Selected character-priced families include monthly free usage, but you must enable billing and pay for excess usage. Standard and WaveNet share the same pricing SKU and listed 4-million-character allowance, rather than providing two additive allowances. Neural2, Chirp 3 HD and Studio list 1-million-character allowances. Gemini and Instant Custom Voice have no listed free allowance.

How much does Gemini text-to-speech cost?

Gemini 2.5 Flash lists at USD 0.50 per million text input tokens plus USD 10 per million audio output tokens; Pro lists at USD 1 plus USD 20 respectively. At Google’s published 25 audio tokens per second, one minute of generated audio has an output-only list cost of USD 0.015 for Flash or USD 0.03 for Pro, plus input-token charges. Other models have different rates.

Can it generate a conversation with multiple speakers?

Yes. Gemini 2.5 Flash and Pro, and Preview 3.1 Flash, support multi-speaker speech with a separate voice assigned to each speaker. Preview 2.5 Flash-Lite supports a single speaker. This capability should not be assumed for every stock voice or model family.

Does Chirp 3 HD support SSML and pronunciation controls?

Chirp SSML support is in Preview and applies to synchronous requests; streaming requests do not support those SSML tags. Separate Preview controls provide speaking rate from 0.25 to 2 times normal speed, pause tags in the markup field, and IPA/X-SAMPA word pronunciations. Pause and pronunciation availability varies by locale, and pause duration is not an exact fixed interval.

What are the script and audio limits?

Ordinary Cloud TTS content is limited to 5,000 bytes per request. For Gemini through Cloud TTS, text and prompt fields are each limited to 4,000 bytes, with 8,000 bytes combined; generated audio is limited to approximately 655 seconds and can be truncated beyond that. UTF-8 bytes are not always the same as characters. A separate Preview asynchronous long-audio operation accepts up to 1 million input bytes and writes to Cloud Storage, subject to its own voice and operation support.

Can Google Cloud clone my voice?

Chirp 3 Instant Custom Voice can create a personal voice after sales allowlisting. It requires the prescribed speaker-consent statement and a reference recording, each up to 10 seconds, recorded in the same environment. A custom consent script cannot replace Google’s supplied statement. Language, region and multilingual-transfer support are limited.

Which audio formats work with streaming?

For Gemini through Cloud TTS, unary requests support LINEAR16, PCM, MP3, Ogg Opus, A-law and mu-law. Its streaming output supports PCM, Ogg Opus, A-law and mu-law, with no MP3 option. Gemini through Vertex AI returns raw 16-bit 24 kHz PCM without a WAV header. Chirp and custom voices have separate encoding lists, so a format offered by one route is not automatically available on another.

Can I use Gemini 3.8 speech through the Cloud TTS API?

No. Gemini 3.8 Flash and Flash-Lite speech, including their voice design and replication features, are Preview offerings through the separate Gemini Enterprise API. Cloud Text-to-Speech supports its listed Gemini 2.5 models and Preview 3.1 Flash, with model-specific limits and availability.

Is this your product? Claim this page to update your listing.