
Rivya Journal

Author
Categories
Keep reading
Read more guides and product notes on related topics.
The fastest way to start audio work in Rivya is to name the deliverable before choosing a model. Narration, multi-speaker dialogue, ambience, music, lyrics, and repair are different jobs.
This guide describes the current custom Music generation path alongside available Sounds and Lyrics; follow-up actions have separate eligibility.
Suno Speech — Suno Speech creates spoken audio with tone, rhythm and emotion. Simple mode writes a script from your scene; Advanced mode reads the script you provide.
Choose Simple to describe an announcement, introduction or spoken scene and let the model write it. Choose Advanced when the wording matters: your main text becomes the script, with optional delivery settings. Switching to Simple drops the Advanced settings.
Suno Speech generates expressive spoken audio and can add background music. For a script with one of the named Zephyr, Puck or Kore voices and WAV output, choose an available Gemini TTS model instead.
Gemini 3.8 Flash TTS · Gemini 3.8 Flash Lite TTS · Gemini 3.1 Flash TTS · Gemini 2.5 Pro TTS
| Job | Current first path | Why it fits |
|---|---|---|
| one speaker reading a script | ElevenLabs Turbo 2.5 | a practical starting point for straightforward spoken delivery |
| speech across languages | ElevenLabs Multilingual V2 | designed for multilingual voice generation |
| several speakers in one scene | ElevenLabs Dialogue V3 | built around speaker turns and dialogue structure |
| ambience, loops, or short sound textures | Suno Sounds | the currently available prompt-to-sound path |
| music | Suno Music | Custom generation available |
| lyrics without an audio render | Suno Lyrics | a focused text output for lyric drafting |
| cleanup of an existing recording | Audio Isolation status | currently paused; no cleanup task can run in Rivya |
The model directory shows current availability before you spend credits. ElevenLabs Sound Effect V2 is also paused, so use Suno Sounds for a live ambience or sound-sketch workflow.
Ask three questions:
Is the input text, an uploaded source, or a saved project?
Should the output be speech, dialogue, a sound layer, lyrics, or music?
Must the original recording survive, or is a newly generated replacement acceptable?
If you need to preserve an existing performance, the job is repair. Rivya does not currently offer a runnable cleanup model, and generating replacement audio is not restoration. Read the audio cleanup status guide before changing course.
For one speaker, prepare the final script first and start with a general text-to-speech model such as ElevenLabs Turbo 2.5.
Use ElevenLabs Multilingual V2 when pronunciation and delivery across languages are central to the job. Review names, numbers, abbreviations, and code-switching separately in every target language.
Use ElevenLabs Dialogue V3 when the script is a scene. Label speakers clearly and keep each turn easy to attribute. A dialogue model cannot fix a script whose speakers or intentions are ambiguous.
Suno Sounds is suited to loops, background textures, environmental layers, and short audio sketches. Describe the asset's role, duration, texture, motion, and anything it must avoid.
Set loop behavior only when repetition matters. Add BPM or key when the sound must fit an edit or existing music; otherwise leave those choices broad for the first run.
Suno Music can create a new song or instrumental in Custom mode. Music extension, WAV conversion and separation remain unavailable; use the current form and check its estimate before submitting.
Suno Sounds and Suno Lyrics remain available. Vocal separation is allowed only for eligible successful Sounds results. Existing Music tasks can still be read, polled for status, and downloaded when results exist.
A useful lyric draft is not automatically a finished song, and a sound texture is not a music arrangement.
For a deeper route, read How to Create AI Music with Rivya and Music Workflows in Rivya.
Before generating, write down:
deliverable and intended use
script, prompt, or source asset
language and speaker structure
required duration or pacing
tone, texture, and elements to avoid
what would make the first result worth saving
whether the result may contain private, licensed, or sensitive material
Then check the model page for current inputs, controls, credit guidance, and availability.
For speech, review pronunciation, pacing, speaker consistency, and whether the delivery supports the message. For sound and music, review timing, loop seams, unwanted elements, and whether the asset works in its real context.
If the branch was wrong—dialogue treated as narration, ambience treated as music, or repair treated as generation—switch early. Save the strongest valid result in History before spending credits on more variations.
Compare available audio models.
Read Best AI Voice Generator for voice model choices.
Read Best AI Sound Generator in Rivya for ambience and sound sketches.
Read AI Audio Cleanup in Rivya for the current repair-tool status.
Use Audio Workflows in Rivya, References and Uploads, and History for product workflow details.