Skip to content

Commit 5ba470b

Browse files
fix: honour default_voice_code in _split_multi_voice (issue remsky#514) (remsky#515)
_split_multi_voice computed segment_lang as the first letter of the voice name when no explicit lang_code was provided, bypassing settings.default_voice_code entirely. This caused voices whose filename does not begin with a valid Kokoro language code to be routed to the wrong pipeline, returning empty audio (HTTP 200 with a zero-byte body). Mirror the precedence used in kokoro_v1.py: lang_code > settings.default_voice_code > voice[:1] Add two unit tests to guard the fix and the existing explicit-wins case. Fixes remsky#514 Signed-off-by: Christian-Sidak <[email protected]>
1 parent 24a5c96 commit 5ba470b

2 files changed

Lines changed: 78 additions & 2 deletions

File tree

‎api/src/services/tts_service.py‎

Lines changed: 6 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -291,8 +291,12 @@ async def _split_multi_voice(
291291
resolved[segment_voice] = await self.get_voices_path(segment_voice)
292292
voice_name, voice_path = resolved[segment_voice]
293293

294-
# request lang_code wins, else each speaker gets the pipeline their prefix implies
295-
segment_lang = lang_code if lang_code else segment_voice[:1].lower()
294+
# request lang_code wins, then default_voice_code setting, else each speaker gets the pipeline their prefix implies
295+
segment_lang = (
296+
lang_code
297+
or settings.default_voice_code
298+
or segment_voice[:1].lower()
299+
)
296300
logger.debug(
297301
f"Using voice path '{voice_path}' with lang_code '{segment_lang}'"
298302
)

‎api/tests/test_tts_service.py‎

Lines changed: 72 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -223,6 +223,78 @@ async def test_split_multi_voice_lang_code_per_speaker():
223223
assert langs == ["a", "b"]
224224

225225

226+
@pytest.mark.asyncio
227+
async def test_split_multi_voice_default_voice_code_used_when_no_lang_code():
228+
"""settings.default_voice_code is used when no explicit lang_code is given.
229+
230+
Regression test for issue #514: voices whose filename does not begin with a
231+
recognised language code were silently matched against the wrong pipeline
232+
(or produced empty audio) because default_voice_code was never consulted.
233+
"""
234+
model_manager = AsyncMock()
235+
voice_manager = AsyncMock()
236+
237+
with (
238+
patch("api.src.services.tts_service.get_model_manager") as mock_get_model,
239+
patch("api.src.services.tts_service.get_voice_manager") as mock_get_voice,
240+
patch("api.src.services.tts_service.settings") as mock_settings,
241+
):
242+
mock_get_model.return_value = model_manager
243+
mock_get_voice.return_value = voice_manager
244+
mock_settings.default_voice_code = "a"
245+
mock_settings.voice_weight_normalization = True
246+
247+
service = await TTSService.create()
248+
# "ursa" does not start with a valid lang prefix; without the fix it
249+
# resolves to "u", which is not a Kokoro pipeline and yields empty audio.
250+
service.get_voices_path = AsyncMock(
251+
side_effect=lambda voice: (voice, f"/path/to/{voice}.pt")
252+
)
253+
254+
langs = [
255+
lang
256+
async for _name, _path, lang, _rate, _text, _tokens, _pause in (
257+
service._split_multi_voice("Hello.", "ursa", None, None)
258+
)
259+
]
260+
261+
assert langs == ["a"], (
262+
"default_voice_code should be used when lang_code is absent and the "
263+
"voice name has no recognised language prefix"
264+
)
265+
266+
267+
@pytest.mark.asyncio
268+
async def test_split_multi_voice_explicit_lang_code_beats_default():
269+
"""An explicit request lang_code takes priority over settings.default_voice_code."""
270+
model_manager = AsyncMock()
271+
voice_manager = AsyncMock()
272+
273+
with (
274+
patch("api.src.services.tts_service.get_model_manager") as mock_get_model,
275+
patch("api.src.services.tts_service.get_voice_manager") as mock_get_voice,
276+
patch("api.src.services.tts_service.settings") as mock_settings,
277+
):
278+
mock_get_model.return_value = model_manager
279+
mock_get_voice.return_value = voice_manager
280+
mock_settings.default_voice_code = "a"
281+
mock_settings.voice_weight_normalization = True
282+
283+
service = await TTSService.create()
284+
service.get_voices_path = AsyncMock(
285+
side_effect=lambda voice: (voice, f"/path/to/{voice}.pt")
286+
)
287+
288+
langs = [
289+
lang
290+
async for _name, _path, lang, _rate, _text, _tokens, _pause in (
291+
service._split_multi_voice("Hello.", "ursa", "b", None)
292+
)
293+
]
294+
295+
assert langs == ["b"], "explicit lang_code must win over default_voice_code"
296+
297+
226298
@pytest.mark.asyncio
227299
async def test_split_multi_voice_explicit_lang_code_wins():
228300
"""An explicit request lang_code overrides every speaker's prefix."""

0 commit comments

Comments
 (0)