Spaces:
Sleeping
Sleeping
shashankpandya
fix: combine rate+pitch into single frame-rate pass, tune emotion params, add output pruning
ac53fef Download engine/mapper.py from shashankpandya/empathy-engine: direct link, hf CLI and curl.
- Browser
- Download file 2.12 kB
-
https://huggingface.co/spaces/shashankpandya/empathy-engine/resolve/main/engine/mapper.py
- Command line
-
hf download hf://spaces/shashankpandya/empathy-engine/engine/mapper.py
-
curl -L -o mapper.py https://huggingface.co/spaces/shashankpandya/empathy-engine/resolve/main/engine/mapper.py
2.12 kB
| # BASE_MAP stores absolute target values at full intensity (intensity = 1.0). | |
| # Neutral baseline: rate=1.0, pitch=0, volume_db=0. | |
| # Each param is linearly interpolated from the neutral baseline toward the | |
| # target value, scaled by the model's confidence score (intensity). | |
| # | |
| # Tuning notes: | |
| # - Rate and pitch are now combined in a single frame-rate pass in tts.py, | |
| # so values here represent the true perceptual target without compensation. | |
| # - Volume deltas are kept moderate — gTTS output is already normalised, | |
| # so large boosts clip and large cuts become inaudible. | |
| BASE_MAP: dict[str, dict] = { | |
| "joy": {"rate": 1.25, "pitch": 3, "volume_db": 4}, | |
| "sadness": {"rate": 0.72, "pitch": -3, "volume_db": -4}, | |
| "anger": {"rate": 1.35, "pitch": 2, "volume_db": 6}, | |
| "fear": {"rate": 1.18, "pitch": 4, "volume_db": -3}, | |
| "disgust": {"rate": 0.85, "pitch": -2, "volume_db": -3}, | |
| "surprise": {"rate": 1.38, "pitch": 5, "volume_db": 4}, | |
| "neutral": {"rate": 1.00, "pitch": 0, "volume_db": 0}, | |
| } | |
| EMOTIONS: list[str] = list(BASE_MAP.keys()) | |
| # Neutral baseline used for interpolation | |
| _NEUTRAL = {"rate": 1.0, "pitch": 0, "volume_db": 0} | |
| def get_params(emotion: str, intensity: float) -> dict: | |
| """ | |
| Returns {'rate': float, 'pitch': int, 'volume_db': int}. | |
| Each parameter is linearly interpolated from the neutral baseline toward | |
| the emotion's target value, scaled by intensity (model confidence score). | |
| Example: joy at intensity=0.5 | |
| rate = 1.0 + (1.25 - 1.0) * 0.5 = 1.125 | |
| pitch = round(0 + (3 - 0) * 0.5) = 2 | |
| volume_db = round(0 + (4 - 0) * 0.5) = 2 | |
| Falls back to neutral params for unknown emotion labels. | |
| """ | |
| target = BASE_MAP.get(emotion, BASE_MAP["neutral"]) | |
| return { | |
| "rate": _NEUTRAL["rate"] + (target["rate"] - _NEUTRAL["rate"]) * intensity, | |
| "pitch": round(_NEUTRAL["pitch"] + (target["pitch"] - _NEUTRAL["pitch"]) * intensity), | |
| "volume_db": round(_NEUTRAL["volume_db"] + (target["volume_db"] - _NEUTRAL["volume_db"]) * intensity), | |
| } | |