Text to Speech API

A text to speech API generates spoken audio from written text. It enables narration and voice features in an application without recording every script manually. The selected model determines available voices, languages, speaking controls and output formats.

Test your actual scripts, including names, numbers and long sentences. Compare pronunciation, naturalness, expressive control, voice consistency and cost. Confirm the required language and audio format, and check the documented request flow if response time is important to your product.

Available controls vary by model. An endpoint may offer voice presets, speed, stability, style instructions or audio tags. Use documented fields and supported values; a model without a speed or emotion control should not be expected to interpret an arbitrary parameter.

Language coverage and pronunciation quality depend on the model and voice. Check the documented language options or detection behavior, then test native text and mixed-language passages. A multilingual label does not guarantee equal quality for every accent or proper name.

Rewrite the passage for spoken language, expand ambiguous abbreviations and test a short sentence. Use documented pronunciation or normalization controls where available. Keep a reviewed text version for narration rather than assuming a written display string is always the best speech input.

Check the endpoint text limit and supported duration behavior. Divide long scripts at natural paragraph or sentence boundaries, reuse voice settings and review transitions after assembly. Preserve context fields where the model supports them to help continuity across separate requests.

No. Selecting a synthetic voice or preset is different from creating a clone from a recording. Voice cloning requires a specifically supported workflow and appropriate permission. Check the available endpoint before designing a product around a custom voice identity.

Some models provide timestamps or multiple-speaker controls, while others do not. Confirm the output fields and request options on the model page. Validate alignment on your own script before using timestamps for subtitles or synchronized text highlighting.

Use the formats and quality options documented for the chosen TTS model. Playback, telephony and editing can require different codecs or sample rates. Inspect the completed output and convert it in a separate media step when your application needs an unsupported format.

Use the selected model generation endpoint, save task_id and retrieve the completed audio through status queries or a supported callback_url. Do not infer a streaming endpoint from the phrase low-latency TTS; check the exact PoYo interface documented for that model.

TTS pricing can use characters, tokens, generation or another model-specific unit. Check how the chosen tier counts usage and how optional settings affect it. Estimate the complete script and expected retries, rather than comparing raw prices with different billing units.

Generate and review the speech first, then confirm that its format and length meet the avatar endpoint requirements. Supply the audio with a suitable portrait and track the avatar task separately. Keep narration text and voice settings so revisions can reproduce the intended delivery.