Typecast is a text-to-speech service with a REST API for generating spoken audio from text in apps, content pipelines, and agents. You pick a voice from a library of several hundred built-in voices, or clone a custom voice from a short audio sample, and send text of up to 2,000 characters per request.
Two models are available. ssfm-v30 covers 37 languages and adds context-aware emotion inference and extra emotion presets; ssfm-v21 covers 27 languages with a smaller preset set. Language is auto-detected when you omit the ISO 639-3 code.
Emotion is set through one of three prompt types: a smart prompt that infers delivery from the surrounding previoustext and nexttext, a preset prompt with an emotion and an intensity value, or a basic prompt for the older model. Output is returned as a complete WAV or MP3 file, or streamed for real-time playback, and a timestamps endpoint returns word- or character-level alignment for subtitles and lip-sync.
Features
- Voice library: built-in voices (tc ids) filterable by model, gender, age, and use case, plus cloned custom voices (uc ids)
- Emotion presets: normal, happy, sad, angry, whisper, toneup, and tonedown, each with an intensity from 0.0 to 2.0
- Smart emotion: the ssfm-v30 model infers the fitting emotion from neighboring text context
- Audio controls: volume (0–200), pitch (−12 to +12 semitones), and tempo (0.5×–2.0×)
- Output formats: 16-bit mono 44.1 kHz WAV or 320 kbps MP3, as a file or a stream
- Timestamps: word- or character-level alignment data for subtitles and karaoke-style highlighting
- Deterministic output: an optional 32-bit seed reproduces the same rendering
- SDKs: official libraries for Python, JavaScript, Go, Rust, C#, Java, Kotlin, C, Swift, Zig, PHP, Dart, and Ruby
- Integrations: Pipecat and LlamaIndex service wrappers, and no-code connectors for Zapier, Make, n8n, and Google Sheets
