Doubao Speech Synthesis Small Model (Speech Synthesis)

What This Is

An earlier "small model" product line within Volcano Engine's Doubao Speech, with the official product name simply "Speech Synthesis". Voice IDs start with BV (e.g., Can Can BV700_streaming). These are a completely different set of voices from the large model 1.0/2.0 — not a lower-tier version of the same lineup.

Suitable Scenarios

Price-sensitive, high-volume scenarios — the postpaid unit price is a fraction of the large model 1.0/2.0. The number of voices is also far greater than the large model (the official voice library has over a hundred voices, covering more granular categories such as general, audiobook reading, intelligent assistant, video dubbing, advertising dubbing, news broadcasting, dialects, and multilingual; Soniva currently includes 24 representative voices).

Voices

24 curated voices (Volcano Engine's complete BV voice library has over a hundred). Unlike the large model 1.0/2.0, this tier's voice list is a built-in static list in Soniva and is not pulled in real time — Volcano Engine's voice list API (ListBigModelTTSTimbres) only covers large model voices and does not return BV voices, so the number here is fixed.

Coverage:

• General scenarios: Can Can, General Female/Male Voice, Ultra-natural Voice - Zi Zi
• Audiobook reading: Qing Cang, Sunny Youth, Commanding Mature Male, Gentle Lady
• Intelligent assistant: Sweet Xiao Yuan, Intellectual Female Voice, Friendly Male Voice
• Video dubbing: Dubbed Film Male Voice, Energetic Commentary Male, Live-streaming Queen
• Featured voices: Adorable Toddler, Child Prodigy
• Advertising dubbing: Promo Male Voice, Magnetic Male Voice
• News broadcasting: News Female Voice, News Male Voice
• Fun dialects: Northeastern Bro, HK Drama Heartthrob
• Multilingual: Lazy Female Voice - Ava (American English), Genius Girl (supports 8 languages)

Pricing

Note that the billing unit is different from the large model — the large model is billed by "character count", while the small model's "Speech Synthesis" is billed by "call count", with a maximum text limit of approximately 300 Chinese characters per call. For short-text, high-frequency scenarios, it is usually much cheaper than the large model.

Technical Limitations (Internal Note — Not to Be Included in Help Center Copy)

These voices currently can only use Volcano Engine's legacy App ID + Access Token authentication; the new single-key (V3) API does not support them. Soniva has internally reconnected the legacy authentication logic specifically to serve this tier. From the user experience perspective, it is the same as 1.0/2.0 — just select a model and a voice, with no perceptible difference in the underlying authentication method.

How to Use in Soniva

TTS Studio → Select "Doubao Speech" as the provider → Select "doubao-tts-classic" (small model) as the model, and the voice list will automatically switch to these BV voices.

Not Currently Supported: Premium Long-Text Speech Synthesis

Under Volcano Engine's small model, there is also a "Premium Long-Text Speech Synthesis" product (asynchronous submission + callback). Its voice library is identical to "Speech Synthesis", but it is designed for very long text and uses asynchronous callbacks (callback_url) instead of synchronous requests. Soniva has not yet integrated this product — the current "doubao-tts-classic" uses the synchronous API, with a maximum text limit of approximately 300 characters per call, which is sufficient for most scenarios. If support for longer one-shot synthesis is needed in the future, additional development of an asynchronous callback pipeline would be required. This capability should not be reflected in the Help Center for now.