starter.wiki

AI

Text-to-Speech

Choose a speech-generation API by realtime behavior, voice fit, language, controllability, consent and rights, data policy, and total delivery cost.

Recommendation

Choose the voice workflow and rights boundary first.

Define realtime versus batch, language, pronunciation, voice source, consent, cloning, disclosure, output format, latency, and human listening criteria before comparing providers.12

For: Application teams adding narration, accessibility, conversational, support, media, or voice-agent speech through a managed API

Main trade-off

Managed TTS reduces synthesis infrastructure while voice rights, cloning and consent, model and catalog lifecycle, latency, streaming, data handling, output policy, and character or audio pricing remain provider dependencies.1234

Choose the product and operating model first

Start from workload, integration, quality, policy, data, lifecycle, and ownership requirements before comparing product breadth.

  1. Voice workload

    Test real scripts, languages, names, numbers, pronunciation, emotion, long-form continuity, interruptions, and accessibility requirements.34

  2. Realtime behavior

    Measure time to first audio, streaming cadence, cancellation, buffering, WebSocket lifecycle, telephony formats, and recovery.2

  3. Rights and policy

    Verify voice source, speaker consent, cloning entitlement, disclosure, prohibited use, output terms, moderation, data use, retention, and deletion.12

  4. Operations and cost

    Verify voice and model lifecycle, language and feature support, character or audio billing, concurrency, region, quotas, versioning, and migration.12

Decision routes

Each route addresses a distinct workload or ownership boundary and retains an explicit verification condition.

OpenAI speech generation

Choose OpenAI speech generation

Choose OpenAI when its current speech generation fits the workload and an existing OpenAI platform relationship is deliberate.

Verify: Verify model and endpoint, voices, languages, streaming, disclosure and policy obligations, output format, data controls, limits, latency, and price.1

Realtime voice API

Choose Realtime voice API

Choose Deepgram when streaming conversational speech, WebSocket delivery, telephony formats, or a shared speech platform boundary are material.

Verify: Verify model and voice identity, languages, continuous streaming, encoding, dedicated endpoints, model-improvement policy, limits, connection behavior, and price.2

Voice quality and catalog

Choose Voice quality and catalog

Choose ElevenLabs when its current voice catalog, expressive models, languages, long-form behavior, or authorized cloning routes fit the product.

Verify: Verify voice and model availability, catalog API access, consent and cloning policy, disclosure, retention mode, output rights, format, limits, and pricing.3

Low-latency voice

Choose Low-latency voice

Choose Cartesia when a specialized low-latency Sonic API, streaming behavior, and supported voice controls match the workload.

Verify: Use current Sonic models and Voice IDs; verify deprecated models and endpoints, version headers, language, controls, cloning and consent, latency, limits, and pricing.4

Boundary: Text-to-Speech owns synthesis from text, realtime output, voices, language, pronunciation, controllability, rights, consent, and delivery behavior. Speech-to-Text owns recognition of input audio.

Differences that change the choice

Compare the workload, product boundary, policy, lifecycle, data path, cost shape, and operating ownership rather than feature volume.

Voice fit
Catalog breadth, expression, pronunciation, language, long-form consistency, and authorized cloning differ by model and voice.13
Realtime delivery
REST, streaming, WebSocket, cancellation, audio formats, and first-byte behavior determine conversational fit.2
Rights and consent
Voice source, cloning permission, disclosure, prohibited use, and output terms are separate from acoustic quality.12
Lifecycle
Voices, models, snapshots, endpoints, languages, and pricing units can change independently.13

Official resources

Verify current model, API, SDK, product, pricing, policy, data, region, lifecycle, and operating boundaries in first-party material.

Sources

Official documentation supports current product boundaries and verification points; route selection remains a bounded editorial judgment.

  1. 1
    OpenAI speech generation official documentation

    OpenAI · Accessed Official

  2. 2
    Deepgram official documentation

    Deepgram · Accessed Official

  3. 3
    ElevenLabs official documentation

    ElevenLabs · Accessed Official

  4. 4
    Cartesia official documentation

    Cartesia · Accessed Official

  5. 5
    Cartesia TTS API changes

    Cartesia · Accessed Official

  6. 6
    ElevenLabs text-to-speech API

    ElevenLabs · Accessed Official