Spark TTS

Spark TTS TTS

Voice cloning from five seconds of audio combined with prompt-based control over emotion, speed, and speaking style.

Ojejapo 5000 caracter rehegua límite

Ojehaijey ñe'ẽnguéra etiquetas SSML-pe peteĩ control hekopete g̃uarã:

<speak><prosody rate="slow">Slow speech</prosody></speak>

Etiquetas ohechakuaáva modelo ojeporavóva - tesãirã peteĩ peteĩva ñe'ẽnguérape, oĩhápe:

Ko modelo ohai texto ndahasyivéva, upévare umi etiqueta oĩva línea ryepýpe ndojehechakuaái. Umi emoción oñemopyendáva etiqueta-pe g̃uarã, oñemoambue peteĩ modelo expresivo-pe taha'e Orfeo térã Bark.

Oñemohenda ñe'ẽnguéra ojehechapyréva (tembiapo = ñe'ẽnguéra):

-12 +12
0.5x 2.0x
Libre Piper, VITS, MeloTTS ndive
Audio-kuéra oguenohẽva ojekuaauka ko'ápe. Oñeporavo peteĩ modelo, omoĩnge ñe'ẽ ha ohesa'ỹijo Generar.
Audio oñemoheñói porã
0:00
Oñeguenohẽ marandu myambue guive Oñeguenohẽ.srt Ko enlace hi'are 24 h rire
Nivel libre: jeiporu personal. Licencia comercial $5/ha'e rupi
Ehayhuetéva TTS.ai? He'i umi iñangirũpe!

Mba'épa Spark TTS

Spark TTS by SparkAudio merges voice cloning with controllable delivery in a single prompt-driven system. Using just five seconds of reference audio it clones a voice, then lets you steer emotion, speed, and speaking style while keeping that cloned identity intact. Under the hood it combines a BiCodec audio tokenizer, an LLM, and flow matching, and it supports English and Chinese. It is aimed at content creation where a single cloned voice needs to express a range of moods and pacing. Note the licensing split: Spark's code is Apache 2.0, but the model weights are released under CC BY-NC-SA 4.0, which restricts commercial use.

Oñeha'ãvéva: Content creation with cloned voices and emotional control

Ojehecha opavave Spark TTS ñe'ẽ

Peteĩ jehecha

Desarrollador
SparkAudio
Licencia
CC BY-NC-SA 4.0
Ta'ãnga
standard
Velocidad
medium
Clonación ñe'ẽnguéra rehe
Ha'e
Ñe'ẽ
English, Chinese
Caracteres máx.
1000

Spark TTS ñe'ẽ

Chinese Default

Chinese
Estándar Neutral

Default

English
Estándar Neutral

Spark TTS Pregunta ojehechavéva

It uses a prompt-based control system layered on top of voice cloning, so you can adjust emotion, speed, and speaking style while preserving the identity of the cloned voice.

About five seconds of reference audio is enough to clone a voice in English or Chinese.

Its model weights are licensed CC BY-NC-SA 4.0, which prohibits commercial use, even though the project code is Apache 2.0. Choose a permissively-licensed model for commercial work.
← Opaite ñe'ẽ