kokoro: persist custom voices across container recreate
Wrapper only enumerates one voice directory (settings.voices_dir, default /app/api/src/voices/v1_0 — inside the container's writable layer, not bind-mounted). Override via VOICES_DIR=/app/user_voices (host bind mount) and add a command shim that cp -r's built-ins from the in-image v1_0 into user_voices on every start. Built-ins re-seed fresh from the image (so upgrades that add voices propagate); custom .pt files in user_voices are preserved (cp -r is additive). Also adds scripts/blend_kokoro_voice.py + a playbook around it that mirrors the wrapper's request-time voice="a(w)+b(w)" math but writes the result as a named .pt to user_voices, making it discoverable via GET /v1/audio/voices and persistent across recreate. Defaults to athena = af_bella(2)+af_aoede(1) normalized.
This commit is contained in:
@@ -0,0 +1,78 @@
|
||||
# Blend two or more Kokoro voicepacks into a new persistent voice.
|
||||
#
|
||||
# Mirrors the wrapper's request-time `voice="a(w)+b(w)"` blend op
|
||||
# (api/src/services/tts_service.py::_get_voices_path) — same math,
|
||||
# but the result is written to /worktank/kokoro/user_voices/ on the
|
||||
# host (bind-mounted into the container at /app/user_voices, which
|
||||
# the compose's VOICES_DIR env points the wrapper at). Result:
|
||||
# - immediately appears in GET /v1/audio/voices
|
||||
# - usable as voice="<out_name>" in /v1/audio/speech
|
||||
# - persistent across `up --force-recreate` and image upgrade
|
||||
# (it's on the host bind mount, not the container layer)
|
||||
#
|
||||
# Requires the persistent-voices compose layout (entrypoint shim +
|
||||
# VOICES_DIR=/app/user_voices). If the deployed compose predates that,
|
||||
# run `scripts/elway irv-ml1 --playbook playbooks/deploy-kokoro.yaml`
|
||||
# first to roll it out.
|
||||
#
|
||||
# Usage (defaults to athena = af_bella(2)+af_aoede(1) normalized):
|
||||
# scripts/elway irv-ml1 --playbook playbooks/blend-kokoro-voice.yaml
|
||||
#
|
||||
# Override per-run:
|
||||
# scripts/elway irv-ml1 --playbook playbooks/blend-kokoro-voice.yaml \
|
||||
# --var 'recipe=af_bella(1)+am_adam(1)' --var out_name=androgyne
|
||||
#
|
||||
# Re-run with --var force=1 to overwrite an existing voice.
|
||||
|
||||
vars:
|
||||
recipe: "af_bella(2)+af_aoede(1)"
|
||||
out_name: athena
|
||||
user_voices_dir: /worktank/kokoro/user_voices
|
||||
container: kokoro
|
||||
script_local: scripts/blend_kokoro_voice.py
|
||||
script_host: /tmp/blend_kokoro_voice.py
|
||||
script_in_container: /tmp/blend_kokoro_voice.py
|
||||
force: "0"
|
||||
api_port: "8193"
|
||||
|
||||
steps:
|
||||
- name: Upload blend script to host /tmp
|
||||
upload:
|
||||
src: "{{ script_local }}"
|
||||
dest: "{{ script_host }}"
|
||||
mode: "0755"
|
||||
|
||||
- name: Copy script into kokoro container
|
||||
shell: docker cp {{ script_host }} {{ container }}:{{ script_in_container }}
|
||||
|
||||
- name: Run blend inside container
|
||||
# `-u 0`: host /worktank/kokoro/user_voices is chowned to 1001:1001
|
||||
# (appuser) by deploy-kokoro.yaml; we still write as root so the
|
||||
# output is plainly 0644 root:root and the wrapper (running as
|
||||
# appuser) can read it. Also avoids any UID-mapping subtlety.
|
||||
# --force is wired through a shell switch so we can stay inside
|
||||
# elway's plain {{ var }} substitution (no jinja conditionals).
|
||||
shell: |
|
||||
FORCE_FLAG=""
|
||||
if [ "{{ force }}" = "1" ]; then FORCE_FLAG="--force"; fi
|
||||
docker exec -u 0 {{ container }} python {{ script_in_container }} \
|
||||
--recipe '{{ recipe }}' --out '{{ out_name }}' $FORCE_FLAG
|
||||
when: "[ ! -f {{ user_voices_dir }}/{{ out_name }}.pt ] || [ '{{ force }}' = '1' ]"
|
||||
|
||||
verify:
|
||||
- name: Output .pt exists on host (persistent)
|
||||
shell: test -s {{ user_voices_dir }}/{{ out_name }}.pt
|
||||
changed_when: "false"
|
||||
|
||||
- name: Voice appears in /v1/audio/voices
|
||||
shell: curl -sf http://localhost:{{ api_port }}/v1/audio/voices | grep -q '"{{ out_name }}"'
|
||||
changed_when: "false"
|
||||
|
||||
- name: Synthesize a probe sample (no playback)
|
||||
shell: |
|
||||
curl -sf -X POST http://localhost:{{ api_port }}/v1/audio/speech \
|
||||
-H 'Content-Type: application/json' \
|
||||
-d '{"model":"kokoro","input":"Hello, I am {{ out_name }}.","voice":"{{ out_name }}","response_format":"wav"}' \
|
||||
-o /tmp/{{ out_name }}-probe.wav \
|
||||
&& test -s /tmp/{{ out_name }}-probe.wav
|
||||
changed_when: "false"
|
||||
Reference in New Issue
Block a user