Skip to content

About

Changeling Echo: train the voices the Changelings speak with - RVC models (female, male, or your own) from Piper speech or your recordings - and speak in them live in Imago Realtime and Mosaic Mask.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Twitter: @NorowaretaGemu License: MIT

Changeling

Changeling Echo

Website: https://cursedprograms.github.io/Changeling-Echo/

The name. Echo was a mountain nymph in Greek myth. Hera, kept talking by her while Zeus slipped away, cursed her to say nothing of her own, only to repeat the last words of others. She faded away for love of Narcissus until nothing was left of her but her voice. Changeling Echo is that voice: someone else's, saying your words.

Part of the Changeling family, all together in Changeling Studio: Imago · Imago Realtime · Tiresias (gender) · Chronos (age) · Natio (where in the world) · Cosmeta (makeup, hair, beards) · Echo (voices) · Tonsor (hairstyles) · realtime: Tiresias · Chronos · Natio · Cosmeta.

Train the voices the Changelings speak with. Echo turns about 15 minutes of one person's speech into an RVC voice model; Imago Realtime and Mosaic Mask then turn your voice into that one, live. The speech can be a built-in voice (Piper's Amy for a woman, Ryan for a man, reading 100 sentences in three deliveries) or your own recordings.

Named after Echo, the nymph of Greek myth who was cursed to speak only in other people's voices.

Use it

Start run.bat (or .\run.ps1, ./run.sh). The first time it sets itself up: a Python 3.12 venv, Applio (cloned; its training engine with PyTorch for CUDA, a few GB) and the two Piper voices. An NVIDIA GPU is needed for training (6 GB is enough).

Changeling Echo's window: the steps from the voice to sending it to the apps

  1. The voice: Female (Amy), Male (Ryan), or My recordings (a folder of WAVs: one speaker, no music, 10 minutes or more). Give it a name (echo_female by default).
  2. The speech: for a built-in voice, Make the speech: Piper reads the 100 sentences in echo/corpus.txt steady, quicker and livelier, and slower and calmer (about 15 minutes of speech, in under a minute).
  3. The models, once: HiFi-GAN (generator and discriminator), RMVPE (pitch) and ContentVec (sound), about 740 MB: only what training at 40 kHz needs, not Applio's whole 2 GB set. A copy already on the PC is used first (F:\AIModels\rvc, or the folders in ECHO_MODEL_DIRS).
  4. Train: slicing, reading pitch and sound, training (a completion bar with the epoch and the time left), indexing. About 300 epochs makes a good voice: an hour or two on a 6 GB card.
  5. Try it: pick a recording and hear it in the new voice.
  6. Use it live: Imago Realtime's Echo switch (and --echo) speaks female and male in echo_female and echo_male. It starts echo/live.py here, in Echo's own environment, and pipes the microphone through it a quarter second at a time: about 0.3 s behind you, under 100 ms of work per quarter second on a 6 GB card.

Unattended: python echo/train_both.py 300 gets the models, makes the speech and trains both built-in voices one after the other. Here that took about 100 minutes each on a 6 GB card. Measured on a test clip:

  • the male reader through echo_female (+12 semitones) went from 158 to 301 Hz;
  • the female reader through echo_male (-12) went from 195 to 97 Hz.

Both voices are also kept in F:\AIModels vc�oices.

How it works

echo/dataset.py makes the speech with Piper (the same one DREAM speaks with; it loads the voice once). echo/train.py runs Applio's own command line for each stage (preprocess, extract, train, index) and copies the newest weights and index to voices/. echo/models.py fetches only the four model files training needs.

Licenses

Echo is MIT; Applio is MIT; Piper's Amy and Ryan voices are from rhasspy/piper-voices. Train voices only from speech you have the right to use: your own, the built-in voices, or someone who agreed.



© Cursed Entertainment 2026


About

Changeling Echo: train the voices the Changelings speak with - RVC models (female, male, or your own) from Piper speech or your recordings - and speak in them live in Imago Realtime and Mosaic Mask.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Sponsor this project

Packages

Contributors

Languages