File size: 3,149 Bytes
c202fc8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 | """Synthetic-rating alignment per TTS voice (Table 12).
python analysis/06_voice_study/06_synth_alignment.py
For each (voice, synthetic rater): per-clip mean synthetic rating against
per-clip mean human rating, over the 60 clips of the 30 voice-study prompts.
Pearson r, Spearman rho, MAE, and mean signed difference (synthetic − human;
positive = synthetic rater above human).
Human means come from the analyzed rater set (attention-check exclusions
applied): the new voice arms from long_voice_normalized.csv, the af_heart
baseline from the main study's long_audio_table_normalized.csv restricted to
the same 30 prompts. Synthetic means average the 10 draws per clip; the
baseline draws are the main study's runs (data/main_study/synth_long_audio_*), the
new voice arms the re-rendered runs in data/voice_study/synth/.
"""
from __future__ import annotations
from pathlib import Path
import pandas as pd
from scipy import stats
ROOT = Path(__file__).resolve().parent.parent.parent
VOICE_DATA = ROOT / "data/voice_study"
CONTEXT_DATA = ROOT / "data/audio_context_study"
DERIVED = ROOT / "results/06_voice_study"
SD = ROOT / "data/main_study"
MODELS = ["gpt4o", "gemini_36_flash", "qwen35_omni_flash"]
ARMS = ["kokoro_am_michael", "chatterbox", "qwen3tts_serena"]
def clip_means(df: pd.DataFrame) -> pd.Series:
return df.groupby("AudioClip")["Rating"].mean()
def align(h: pd.Series, s: pd.Series) -> dict | None:
m = pd.concat([h.rename("h"), s.rename("s")], axis=1).dropna()
if len(m) < 3:
return None
return dict(clips=len(m),
r=stats.pearsonr(m.h, m.s)[0],
rho=stats.spearmanr(m.h, m.s)[0],
mae=(m.s - m.h).abs().mean(),
signed=(m.s - m.h).mean())
def main() -> None:
prompts = set(pd.read_csv(ROOT / "stimuli/voice_study/prompts.csv")["PromptID"])
human = {}
voice = pd.read_csv(VOICE_DATA / "long_voice_normalized.csv")
for arm, g in voice.groupby("Arm"):
human[arm] = clip_means(g)
base = pd.read_csv(SD / "long_audio_table_normalized.csv")
base = base[~base["IsAttentionCheck"].astype(bool)]
base = base[base["PromptID"].isin(prompts)]
human["kokoro_af_heart (baseline)"] = clip_means(base)
rows = []
for voice_name, h in human.items():
for model in MODELS:
if "baseline" in voice_name:
path = SD / f"synth_long_audio_table_{model}.csv"
else:
path = (VOICE_DATA / "synth" /
f"synth_long_audio_table_{voice_name}_{model}.csv")
if not path.exists():
continue
s = pd.read_csv(path)
s = s[s["PromptID"].astype(str).isin(prompts)]
res = align(h, clip_means(s))
if res:
rows.append(dict(voice=voice_name, rater=model, **res))
out = pd.DataFrame(rows)
print(out.to_string(index=False, float_format=lambda v: f"{v:.3f}"))
out.to_csv(DERIVED / "synth_alignment_by_voice.csv", index=False)
print(f"\n -> results/06_voice_study/synth_alignment_by_voice.csv")
if __name__ == "__main__":
main()
|