thaloco/aman
1
Fork 0
Code Pull requests Releases Packages Activity Actions

Add vocabulary correction pipeline and example config

Browse source
This commit is contained in:
Thales Maciel 2026-02-25 10:03:32 -03:00
parent f9224621fa
commit c3503fbbde
9 changed files with 865 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

107
tests/test_leld.py
View file

@ -11,7 +11,7 @@ if str(SRC) not in sys.path:
sys.path.insert(0, str(SRC))
import leld
from config import Config
from config import Config, VocabularyReplacement
class FakeDesktop:
@ -32,8 +32,43 @@ class FakeSegment:
class FakeModel:
def __init__(self, text: str = "hello world"):
self.text = text
self.last_kwargs = {}
def transcribe(self, _audio, language=None, vad_filter=None):
return [FakeSegment("hello world")], {"language": language, "vad_filter": vad_filter}
self.last_kwargs = {
"language": language,
"vad_filter": vad_filter,
}
return [FakeSegment(self.text)], self.last_kwargs
class FakeHintModel:
def __init__(self, text: str = "hello world"):
self.text = text
self.last_kwargs = {}
def transcribe(
self,
_audio,
language=None,
vad_filter=None,
hotwords=None,
initial_prompt=None,
):
self.last_kwargs = {
"language": language,
"vad_filter": vad_filter,
"hotwords": hotwords,
"initial_prompt": initial_prompt,
}
return [FakeSegment(self.text)], self.last_kwargs
class FakeAIProcessor:
def process(self, text, lang="en", **_kwargs):
return text
class FakeAudio:
@ -48,12 +83,13 @@ class DaemonTests(unittest.TestCase):
cfg.logging.log_transcript = False
return cfg
@patch("leld._build_whisper_model", return_value=FakeModel())
@patch("leld.stop_audio_recording", return_value=FakeAudio(8))
@patch("leld.start_audio_recording", return_value=(object(), object()))
def test_toggle_start_stop_injects_text(self, _start_mock, _stop_mock, _model_mock):
def test_toggle_start_stop_injects_text(self, _start_mock, _stop_mock):
desktop = FakeDesktop()
daemon = leld.Daemon(self._config(), desktop, verbose=False)
with patch("leld._build_whisper_model", return_value=FakeModel()):
daemon = leld.Daemon(self._config(), desktop, verbose=False)
daemon.ai_processor = FakeAIProcessor()
daemon._start_stop_worker = (
lambda stream, record, trigger, process_audio: daemon._stop_and_process(
stream, record, trigger, process_audio
@ -68,12 +104,13 @@ class DaemonTests(unittest.TestCase):
self.assertEqual(daemon.get_state(), leld.State.IDLE)
self.assertEqual(desktop.inject_calls, [("hello world", "clipboard")])
@patch("leld._build_whisper_model", return_value=FakeModel())
@patch("leld.stop_audio_recording", return_value=FakeAudio(8))
@patch("leld.start_audio_recording", return_value=(object(), object()))
def test_shutdown_stops_recording_without_injection(self, _start_mock, _stop_mock, _model_mock):
def test_shutdown_stops_recording_without_injection(self, _start_mock, _stop_mock):
desktop = FakeDesktop()
daemon = leld.Daemon(self._config(), desktop, verbose=False)
with patch("leld._build_whisper_model", return_value=FakeModel()):
daemon = leld.Daemon(self._config(), desktop, verbose=False)
daemon.ai_processor = FakeAIProcessor()
daemon._start_stop_worker = (
lambda stream, record, trigger, process_audio: daemon._stop_and_process(
stream, record, trigger, process_audio
@ -87,6 +124,60 @@ class DaemonTests(unittest.TestCase):
self.assertEqual(daemon.get_state(), leld.State.IDLE)
self.assertEqual(desktop.inject_calls, [])
@patch("leld.stop_audio_recording", return_value=FakeAudio(8))
@patch("leld.start_audio_recording", return_value=(object(), object()))
def test_dictionary_replacement_applies_after_ai(self, _start_mock, _stop_mock):
desktop = FakeDesktop()
model = FakeModel(text="good morning martha")
cfg = self._config()
cfg.vocabulary.replacements = [VocabularyReplacement(source="Martha", target="Marta")]
with patch("leld._build_whisper_model", return_value=model):
daemon = leld.Daemon(cfg, desktop, verbose=False)
daemon.ai_processor = FakeAIProcessor()
daemon._start_stop_worker = (
lambda stream, record, trigger, process_audio: daemon._stop_and_process(
stream, record, trigger, process_audio
)
)
daemon.toggle()
daemon.toggle()
self.assertEqual(desktop.inject_calls, [("good morning Marta", "clipboard")])
def test_transcribe_skips_hints_when_model_does_not_support_them(self):
desktop = FakeDesktop()
model = FakeModel(text="hello")
cfg = self._config()
cfg.vocabulary.terms = ["Docker", "Systemd"]
with patch("leld._build_whisper_model", return_value=model):
daemon = leld.Daemon(cfg, desktop, verbose=False)
result = daemon._transcribe(object())
self.assertEqual(result, "hello")
self.assertNotIn("hotwords", model.last_kwargs)
self.assertNotIn("initial_prompt", model.last_kwargs)
def test_transcribe_applies_hints_when_model_supports_them(self):
desktop = FakeDesktop()
model = FakeHintModel(text="hello")
cfg = self._config()
cfg.vocabulary.terms = ["Systemd"]
cfg.vocabulary.replacements = [VocabularyReplacement(source="docker", target="Docker")]
with patch("leld._build_whisper_model", return_value=model):
daemon = leld.Daemon(cfg, desktop, verbose=False)
result = daemon._transcribe(object())
self.assertEqual(result, "hello")
self.assertIn("Docker", model.last_kwargs["hotwords"])
self.assertIn("Systemd", model.last_kwargs["hotwords"])
self.assertIn("Preferred vocabulary", model.last_kwargs["initial_prompt"])
class LockTests(unittest.TestCase):
def test_lock_rejects_second_instance(self):