Generate
REST and durable URL
Wait for a finished result and receive a URL your application can retain or distribute.
Text-to-Speech model
Loly 3.5: Give your product a voice worth listening to
A product voice is more than an audio file. It is how reassurance sounds, how guidance keeps its rhythm and how a digital experience becomes recognisable with the screen turned off.
On this page
The contract in one screen
Every value below is read from the same modules the public routes are built on, so this table and an endpoint page can never describe two different products.
- Modality
- Text to speech
- Transports
- REST, SSE, WebSocket
- Output formats
- WAV, MP3, PCM16
- Stream frames
- PCM16 mono, 24,000 Hz
- Languages
- 646 plus auto
- Playback speed
- 0.5 - 1.5
- Text per request
- 5,000 characters
- Text per stream token
- 10,000 characters
A voice with continuity
Loly 3.5 turns text into speech while keeping the integration boundary explicit. Your application chooses the words, the voice and the transport. The model carries that intent into audio that can be stored, downloaded or played as it arrives.
Write the meaning once. Deliver it in the rhythm your experience requires.
Choose how the voice arrives
The same model serves both deliberate, file-based moments and live interfaces. Select the transport by what the listener should experience next.
Bytes
REST and binary body
Receive the completed WAV or MP3 directly when your server wants to own the file immediately.
SSE
Progressive PCM16 events
Read base64 audio events from one POST response as synthesis progresses.
WebSocket
Binary PCM16 frames
Open a token-authorised session and play raw mono frames as they arrive.
Shape the delivery without hiding the contract
The controls are explicit request fields, not a hidden style layer. Choose a language, output container and pace, then keep advanced synthesis controls at their defaults until your listening tests give you a reason to change them.
Language
auto, code or full nameUse automatic handling or select from the published Loly 3.5 language catalog.
Format
wav or mp3Choose a PCM container for processing or compressed audio for delivery.
Speed
0.5 to 1.5Set the playback pace within the range enforced by the public contract.
Synthesis
cfg and diffusion stepsAdvanced fields remain available when an integration needs deliberate tuning.
One credential, one voice identity
A voice key is bound to exactly one voice. That boundary makes voice selection visible in your credential design instead of letting client payloads silently switch identity. Account keys can select an allowed voice explicitly when their permissions require it.
A recognisable voice begins with permission
Custom voice enrollment is only for audio whose speaker authorised cloning. The public contract records that consent, keeps ownership scoped to the account and never returns the secret of an existing key a second time.
Know exactly what reaches the player
Finished responses use the selected WAV or MP3 container. SSE carries base64 PCM16 little-endian mono. WebSocket binary frames carry raw PCM16 mono at the sample rate reported by the server, with JSON reserved for control events.
- WAV or MP3 for finished output
- Base64 PCM16 in SSE audio events
- Raw PCM16 in WebSocket binary frames
- Server-reported sample rate and event metadata
Let the first sentence be real
Open the Bytes endpoint for the shortest path to a playable file, or begin with WebSocket when the experience must speak while the moment is still unfolding.
Was this page helpful?