serinity / tests /unit /test_speech_to_text.py
Amogh1221's picture
Upload 137 files
df55c32 verified
Raw History Blame Contribute Delete
4.07 kB
import pytest
from unittest.mock import MagicMock, patch
from providers.speech_to_text_provider import _parse_sensevoice_output, SenseVoiceSTTProvider
# ────────────── Pure parsing function (no model needed) ──────────────────
class TestParseSenseVoiceOutput:
"""Tests for the pure _parse_sensevoice_output helper, no mocks needed."""
def test_extracts_emotion_happy(self):
raw = "<|HAPPY|><|Speech|>Today is a great day."
text, emotion, event = _parse_sensevoice_output(raw)
assert text == "Today is a great day."
assert emotion == "happy"
assert event == "Speech"
def test_extracts_emotion_sad(self):
raw = "<|SAD|>I feel terrible."
text, emotion, event = _parse_sensevoice_output(raw)
assert text == "I feel terrible."
assert emotion == "sad"
assert event is None
def test_defaults_to_neutral_for_no_tag(self):
raw = "Just a plain sentence."
text, emotion, event = _parse_sensevoice_output(raw)
assert text == "Just a plain sentence."
assert emotion == "neutral"
assert event is None
def test_strips_multiple_tags(self):
raw = "<|NEUTRAL|><|Laughter|>Ha ha ha."
text, emotion, event = _parse_sensevoice_output(raw)
assert text == "Ha ha ha."
assert emotion == "neutral"
assert event == "Laughter"
# ──────────────── SenseVoiceSTTProvider (mocked model) ───────────────────
class TestSenseVoiceSTTProvider:
"""Tests for SenseVoiceSTTProvider with FunASR mocked out."""
def test_transcribe_returns_text_and_emotion(self, tmp_path):
"""Provider correctly calls the model and parses the output."""
with patch("providers.speech_to_text_provider.SenseVoiceSTTProvider._init_stt") as mock_init:
provider = SenseVoiceSTTProvider.__new__(SenseVoiceSTTProvider)
provider._sense_voice = MagicMock()
provider._sense_voice.generate.return_value = [
{"text": "<|SAD|><|Speech|>I feel really low today."}
]
result = provider.transcribe(b"fake_audio_bytes")
assert result["text"] == "I feel really low today."
assert result["emotion"] == "sad"
assert result["event"] == "Speech"
def test_transcribe_returns_empty_when_model_not_loaded(self):
"""If SenseVoice didn't load, transcribe gracefully returns empty."""
provider = SenseVoiceSTTProvider.__new__(SenseVoiceSTTProvider)
provider._sense_voice = None
result = provider.transcribe(b"fake_audio_bytes")
assert result["text"] == ""
assert result["emotion"] == "unknown"
def test_transcribe_returns_empty_on_empty_model_response(self):
"""Empty list from FunASR returns safe empty result."""
provider = SenseVoiceSTTProvider.__new__(SenseVoiceSTTProvider)
provider._sense_voice = MagicMock()
provider._sense_voice.generate.return_value = []
result = provider.transcribe(b"fake_audio_bytes")
assert result["text"] == ""
assert result["emotion"] == "neutral"
def test_transcribe_handles_generate_exception(self):
"""If FunASR throws during generate, transcribe returns a safe fallback."""
provider = SenseVoiceSTTProvider.__new__(SenseVoiceSTTProvider)
provider._sense_voice = MagicMock()
provider._sense_voice.generate.side_effect = RuntimeError("Model crash")
result = provider.transcribe(b"fake_audio_bytes")
assert result["text"] == ""
assert result["emotion"] == "unknown"
def test_init_handles_missing_funasr(self):
"""If funasr is not installed, provider initializes gracefully with _sense_voice=None."""
with patch.dict("sys.modules", {"funasr": None, "funasr.utils.postprocess_utils": None}):
provider = SenseVoiceSTTProvider()
assert provider._sense_voice is None