SpaceXAI TTS

Qwen Audio 3.1 TTS Plus

High-quality multilingual speech synthesis for narration, dubbing, brand voices, and character performance. Four preset voices support instruction-guided emotion, tone, and persona, plus SSML and expressive tags. Returns MP3, WAV, or PCM audio after generation completes. Optional API word timestamps return JSON with base64 audio. Billed by measured characters. NanoGPT accepts up to 20,000 characters per request; this is a transport limit.

Voxtral TTS 3

Generate multilingual speech with preset voices or a per-request MP3 or WAV voice reference. Supports inline emotion and pace tags, pronunciation hints, and MP3, WAV, PCM, FLAC, or Opus output. Up to 4,000 input characters; returns completed audio.

Kotoba TTS v2

Japanese speech synthesis with male-01 and female-01 preset voices. Returns completed 24 kHz mono WAV audio. Billed by exact UTF-8 bytes, not characters. NanoGPT accepts up to 12,000 input bytes.

Qwen Audio 3.1 TTS Flash

Speech synthesis with four multilingual voices in Singapore. Supports Japanese, Korean, French, German, Portuguese, Italian, Vietnamese, and Indonesian, plus eight Chinese dialects. Returns MP3, WAV, or PCM audio after generation completes. Billed by measured input and output tokens; character-based previews are estimates.

Gemini 3.8 Flash-Lite TTS

Fast, expressive single-speaker speech in over 100 languages with 30 voices, delivery instructions, and inline vocal events. Accepts text or one narration turn and returns WAV audio. Billed by exact spoken characters with no per-request minimum; instructions are not billed. The complete input is limited to 8,192 tokens; turn-array input also has an 8,192-byte normalized limit.

Gemini 3.8 Flash TTS

Accepts compatible input turns: one turn for narration or exactly two distinct voices for multi-turn dialogue (up to 100 turns and 8,192 normalized input bytes). Expressive narration and two-speaker dialogue with 30 voices, delivery instructions, and inline vocal events. Returns WAV audio. Billed by exact spoken characters with no per-request minimum; instructions are not billed. The complete input is limited to 8,192 tokens.