canonical: https://jentic.com/apis/googleapis.com/texttospeech

# Google Cloud Text-to-Speech API

Cloud Text-to-Speech synthesises natural-sounding speech from text or SSML using Google's neural network voices, including WaveNet, Neural2, and Studio voice tiers. It supports more than 50 languages and locales, dozens of voice variants per language, and SSML controls for pacing, pitch, pauses, and pronunciation. The synchronous synthesize endpoint returns audio for short text; the synthesizeLongAudio endpoint streams long-form audio (audiobooks, course content) to a Cloud Storage bucket via a long-running operation.

## For AI agents

Synthesize natural-sounding speech audio from text or SSML in 50+ languages, including long-form Cloud Storage output and a wide voice catalog.

## Scope

Does not handle speech recognition, audio editing, or voice cloning of arbitrary speakers - use for synthesising speech from text using Google's voice catalog only.

## Capabilities

- Synthesize short-form speech audio from plain text or SSML markup
- Stream long-form audio synthesis output to a Cloud Storage bucket
- List available voices filtered by language and gender
- Apply SSML controls for pacing, pitch, pauses, emphasis, and pronunciation
- Choose between Standard, WaveNet, Neural2, and Studio voice quality tiers
- Track long-running synthesis operations through dedicated operation endpoints

## Use cases

### IVR and Voice Bot Prompts

Contact centres use Cloud Text-to-Speech to generate dynamic prompts for IVR systems and voice bots. Static greetings are pre-rendered with a Studio voice and cached; per-call dynamic text (caller name, account balance) is synthesised on demand using lower-latency Neural2 voices. Output is returned as base64 LINEAR16 audio that integrates with telephony stacks like Dialogflow CX or Twilio.

Example prompt: POST /v1/text:synthesize with input.text='Welcome back, Maria. Your balance is $312.', voice.languageCode='en-US', voice.name='en-US-Neural2-F', and audioConfig.audioEncoding='LINEAR16'.

### Audiobook and Long-Form Narration

Publishers convert book chapters and course transcripts into audio using the synthesizeLongAudio endpoint. The endpoint accepts up to 1 million characters per request, runs as a long-running operation, and writes the resulting audio file directly to a Cloud Storage bucket. This avoids the synchronous endpoint's 5,000-character limit and is the production-grade path for any content longer than a paragraph.

Example prompt: POST /v1/{parent=projects/*/locations/*}:synthesizeLongAudio with the chapter text, parent=projects/PROJECT/locations/us-central1, and outputGcsUri='gs://my-bucket/chapter-1.wav'.

### Multilingual Marketing Voice-Overs

Marketing teams generate localised voice-overs for product videos by sending the same script translated into target languages and selecting a matching Neural2 or Studio voice per locale. This produces consistent brand voice characteristics across markets in minutes rather than scheduling voice talent per language.

Example prompt: Iterate over a list of language-text pairs and POST /v1/text:synthesize for each with voice.languageCode set per locale and voice.name=<lang>-Studio-A.

### Agent-Generated Spoken Responses via Jentic

An AI agent producing voice replies for a smart-home assistant generates the response text from an LLM, then calls Cloud Text-to-Speech through Jentic to render audio. Jentic isolates the GCP credential, returns the audio bytes, and the agent streams them to the smart speaker.

Example prompt: Through Jentic, search 'synthesize speech from text', load POST /v1/text:synthesize, and execute it with the LLM-generated reply text and a chosen Neural2 voice.

## Key endpoints

| Method | Path | Description |
| --- | --- | --- |
| POST | /v1/text:synthesize | Synthesize speech audio from text or SSML |
| GET | /v1/voices | List available voices, optionally filtered by language code |
| POST | /v1/{+parent}:synthesizeLongAudio | Start a long-running synthesis writing to Cloud Storage |
| GET | /v1/{+name}/operations | List long-running synthesis operations |
| POST | /v1/{+name}:cancel | Cancel an in-flight long-running synthesis |

## Key resources

- **Voice** — Catalog of available voices keyed by language code, name, and SSML gender
- **Synthesis** — Synchronous text-to-speech operation returning base64 audio
- **LongAudioSynthesis** — Long-running operation writing synthesised audio to Cloud Storage

## Why Jentic

- **Setup:** Wiring the Cloud Text-to-Speech API by hand means setting up a Google service account, granting the cloud-platform scope, formatting the input, voice, and audioConfig blocks, and tracking long-audio operations yourself. Through Jentic you install once, import the Cloud Text-to-Speech API from the API Directory, store the service-account credential once, and your agent calls it.
- **Permission scoping:** The text to synthesize travels in the request body rather than a resource path, so scope by operation: limit the agent to the operations it needs, such as text:synthesize and listing voices. You choose the operations it may call, so long-audio synthesis is not included unless you add it.
- **Credential handling:** Your Cloud Text-to-Speech service-account credential is stored once, encrypted, by your own Jentic One instance and injected at execution time. It never enters the agent's prompt, logs, or context.
- **Discovery method:** Agents search Jentic by intent such as 'synthesize speech from text' or 'list available voices', and Jentic returns the matching Text-to-Speech v1 operation with its input schema, including input, voice, and audioConfig, so the agent calls the right endpoint without browsing the reference docs.

## Related APIs

- **Cloud Speech-to-Text API** — The reverse direction - transcribe spoken audio into text
- **Cloud Translation API** — Translate source text before synthesising speech in the target language
- **Google Dialogflow API** — Conversational platform that uses Text-to-Speech for spoken agent responses

## FAQ

### What authentication does the Cloud Text-to-Speech API use?

OAuth 2.0 with the cloud-platform scope is required. Most production integrations use a Google Cloud service-account credential and exchange it for a short-lived bearer token. Through Jentic, the service-account JSON is stored encrypted in the vault and Jentic mints scoped tokens per call.

### Can I synthesize long-form audio with the Cloud Text-to-Speech API?

Yes. The synchronous /v1/text:synthesize endpoint caps input at 5,000 characters; for longer content use POST /v1/{parent}:synthesizeLongAudio, which accepts up to 1 million characters per request and writes the resulting audio to a Cloud Storage URI you specify. Track progress via the returned long-running operation.

### What are the rate limits for the Cloud Text-to-Speech API?

The default per-project quota is 1,000 requests per minute for the synchronous synthesize endpoint, with separate character-per-minute caps that vary by voice tier (Standard voices have higher throughput than Studio). Long-audio synthesis allows a smaller number of concurrent operations. Quota increases are available through the Cloud Console.

### How do I generate speech audio through Jentic?

Search Jentic for 'synthesize speech from text', load POST /v1/text:synthesize, and execute it with input.text, voice (languageCode and optional name), and audioConfig.audioEncoding (MP3, LINEAR16, OGG_OPUS). The response includes audioContent as base64. Get started with Jentic One, the self-hosted execution layer.

### What voice tiers and languages are supported?

Standard, WaveNet, Neural2, and Studio voice tiers are available across more than 50 languages. Studio voices offer the highest realism and are recommended for narration; Neural2 balances quality and latency for interactive use; WaveNet is a long-standing high-quality tier; Standard is the cheapest and lowest-latency. List all available voices via GET /v1/voices.

### Is the Cloud Text-to-Speech API free?

Pricing is per-character with a free tier each month. Standard voices and WaveNet/Neural2 voices have separate free-tier allotments; Studio voices have their own pricing. SSML markup is included in the character count. See cloud.google.com/text-to-speech/pricing for current rates.

### Can I limit what my agent is allowed to do with the Cloud Text-to-Speech API?

Yes. Because you run Jentic One yourself, your own rules decide which operations and credentials the agent may use, and the text to synthesize travels in the request body rather than a resource path, so you scope by operation. You can allow the agent only the operations it needs, such as text:synthesize and listing available voices with GET /v1/voices, and leave everything else off. Long-audio synthesis via synthesizeLongAudio, which writes to a Cloud Storage bucket, stays unavailable to the agent unless you explicitly add it. Your service-account credential is injected at execution time and never reaches the agent's prompt or logs.