Spaces:
Sleeping
Sleeping
File size: 23,545 Bytes
a3419b6 29ebcc1 a3419b6 0579ca0 24a256c 0579ca0 a3419b6 29ebcc1 a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 78dff9d a3419b6 29ebcc1 a3419b6 78dff9d 29ebcc1 78dff9d 29ebcc1 78dff9d 29ebcc1 78dff9d 29ebcc1 a3419b6 29ebcc1 78dff9d 29ebcc1 a3419b6 78dff9d a3419b6 78dff9d a3419b6 229a3e3 a3419b6 229a3e3 a3419b6 229a3e3 a3419b6 78dff9d 29ebcc1 a3419b6 78dff9d 29ebcc1 a3419b6 78dff9d a3419b6 29ebcc1 a3419b6 29ebcc1 a3419b6 78dff9d 29ebcc1 a3419b6 29ebcc1 a3419b6 29ebcc1 78dff9d 29ebcc1 a3419b6 | 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 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 | # app/app.py
# Bantrly TTS Evaluation Framework
# Interactive UI for comparing TTS engines across grade bands.
#
# Run from app/ directory:
# uv run gradio app.py
#
# Metrics:
# WER β Word Error Rate (Radford et al. 2023, Whisper)
# UTMOS β Automated MOS prediction (Saeki et al. 2022, VoiceMOS Challenge)
# RTF β Real Time Factor (synthesis_time / audio_duration)
# Cost β Equivalent Chirp 3 HD cost at $16/1M chars
import sys
import os
import tempfile
import pandas as pd
import gradio as gr
from storage import upload_audio_background, download_csv
sys.path.insert(0, os.path.dirname(__file__))
from dotenv import load_dotenv
# loads .env locally β on HF Spaces, secrets are injected as env vars directly
load_dotenv(os.path.join(os.path.dirname(__file__), ".env"), override=False)
from engines import ENGINES, ENGINE_MAP
from engines.kokoro_engine import KOKORO_VOICES, KOKORO_DEFAULT_VOICE
from evaluator import evaluate
from storage import upload_audio_background
from pathlib import Path
# ββ constants βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
BANDS = ["K-2", "3-5", "6-8", "9-12"]
ENGINE_CHOICES = [e.name for e in ENGINES]
_EVAL_LOG_PATH = os.path.join(os.path.dirname(__file__), "results", "eval_log.csv")
# recommended voice per band for Kokoro
KOKORO_BAND_VOICE = {
"K-2": "af_heart",
"3-5": "af_heart",
"6-8": "af_heart",
"9-12": "am_echo",
}
# ββ state βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
_session_results: list[dict] = []
_session_audio_urls: list[str] = []
# ββ helpers βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def format_wer(wer):
if wer is None:
return "N/A"
pct = round(wer * 100, 1)
note = " β (short text)" if wer > 0.5 else ""
return f"{pct}%{note}"
def format_utmos(score):
if score is None:
return "N/A"
return f"{score:.3f} / 5.0"
def format_rtf(rtf):
if rtf is None:
return "N/A"
flag = "β faster than real time" if rtf < 1.0 else "β slower than real time"
return f"{rtf:.3f}x ({flag})"
def format_cost(engine_cost, chirp_cost, engine_name=""):
if "RunPod" in engine_name:
return f"${engine_cost:.6f} (actual)"
if engine_cost == 0.0:
return f"$0.00 (Chirp equiv: ${chirp_cost:.6f})"
return f"${engine_cost:.6f}"
def build_comparison_table(results: list[dict]) -> pd.DataFrame:
columns = [
"Engine",
"Band",
"Voice",
"UTMOS β",
"WER β",
"RTF β",
"Latency (s)",
"Cost",
]
if not results:
return pd.DataFrame(columns=columns)
rows = []
for r in results:
rows.append({
"Engine": r["engine"],
"Band": r["band"],
"Voice": r.get("voice", "β"),
"UTMOS β": format_utmos(r["utmos"]),
"WER β": format_wer(r["wer"]),
"RTF β": format_rtf(r["rtf"]),
"Latency (s)": r["latency_s"],
"Cost": format_cost(r["engine_cost_usd"], r["chirp_equiv_usd"], r["engine"]),
})
return pd.DataFrame(rows)
def build_business_chart(results: list[dict]):
"""
Bubble chart for business decision making.
X = RTF (speed, lower = better)
Y = UTMOS (quality, higher = better)
Bubble size = fixed (cost removed from visual)
Color = engine type
Reads directly from results dicts β no dependency on display column names.
"""
import plotly.graph_objects as go
if not results:
fig = go.Figure()
fig.update_layout(
title="Run a synthesis to see the comparison chart",
height=450,
)
return fig
def parse_rtf(rtf_str):
if rtf_str is None or rtf_str == "N/A":
return None
try:
return float(str(rtf_str).split("x")[0])
except Exception:
return None
def parse_utmos(utmos_str):
if utmos_str is None or utmos_str == "N/A":
return None
try:
return float(str(utmos_str).split(" ")[0])
except Exception:
return None
color_map = {
"neural-local": "#2ecc71",
"neural-cloud-free": "#3498db",
"neural-cloud-paid": "#e74c3c",
"rule-based-local": "#95a5a6",
}
traces = {}
for r in results:
rtf = parse_rtf(format_rtf(r.get("rtf")))
utmos = parse_utmos(format_utmos(r.get("utmos")))
if rtf is None or utmos is None:
continue
engine_name = r["engine"]
engine_type = r.get("engine_type", "neural-local")
voice = r.get("voice", "β")
latency = r.get("latency_s", "β")
wer_str = format_wer(r.get("wer"))
production = "β" if r.get("production_ready") else "β"
color = color_map.get(engine_type, "#bdc3c7")
hover = (
f"<b>{engine_name}</b><br>"
f"Voice: {voice}<br>"
f"UTMOS: {utmos:.3f}<br>"
f"RTF: {rtf:.3f}x<br>"
f"WER: {wer_str}<br>"
f"Latency: {latency}s<br>"
f"Cost: {format_cost(r.get('engine_cost_usd', 0), r.get('chirp_equiv_usd', 0), engine_name)}<br>"
f"Production: {production}"
)
if engine_type not in traces:
traces[engine_type] = {
"x": [], "y": [], "sizes": [],
"hovers": [], "labels": [],
"color": color,
}
traces[engine_type]["x"].append(rtf)
traces[engine_type]["y"].append(utmos)
cost = r.get("engine_cost_usd", 0) or 0
size = 20 + min(cost * 2000, 25)
traces[engine_type]["sizes"].append(size)
traces[engine_type]["hovers"].append(hover)
traces[engine_type]["labels"].append(f"{engine_name}<br>({voice})")
fig = go.Figure()
for engine_type, data in traces.items():
fig.add_trace(go.Scatter(
x=data["x"],
y=data["y"],
mode="markers",
name=engine_type,
showlegend=True,
marker=dict(
size=data["sizes"],
color=data["color"],
opacity=0.85,
line=dict(width=1.5, color="rgba(255,255,255,0.5)"),
),
hovertext=data["hovers"],
hoverinfo="text",
))
fig.add_vline(
x=1.0, line_dash="dash", line_color="rgba(255,255,255,0.4)", opacity=0.8,
annotation_text="RTF = 1.0",
annotation_font_color="rgba(255,255,255,0.7)",
annotation_position="top right",
)
fig.add_hline(
y=4.0, line_dash="dash", line_color="rgba(255,255,255,0.4)", opacity=0.8,
annotation_text="UTMOS = 4.0 threshold",
annotation_font_color="rgba(255,255,255,0.7)",
annotation_position="right",
)
fig.add_annotation(
x=0.1, y=4.9,
text="β Ideal zone<br>(fast + high quality)",
showarrow=False,
font=dict(color="#2ecc71", size=11),
bgcolor="rgba(46,204,113,0.15)",
bordercolor="#2ecc71",
borderwidth=1,
)
all_rtf = [x for t in traces.values() for x in t["x"]]
x_max = max(3.0, max(all_rtf) + 0.5) if all_rtf else 3.0
fig.update_layout(
title=dict(text="TTS Engine Comparison β Business Decision Chart", font=dict(color="white")),
xaxis_title="RTF β (lower = faster synthesis)",
yaxis_title="UTMOS β (higher = more natural)",
height=500,
legend_title="Engine Type",
xaxis=dict(
range=[-0.1, x_max],
color="white",
gridcolor="rgba(255,255,255,0.15)",
title_font=dict(color="white"),
tickfont=dict(color="white"),
),
yaxis=dict(
range=[3.5, 5.0],
color="white",
gridcolor="rgba(255,255,255,0.15)",
title_font=dict(color="white"),
tickfont=dict(color="white"),
),
legend=dict(
title=dict(text="Engine Type", font=dict(color="white", size=12)),
font=dict(color="white"),
bgcolor="rgba(30,30,30,0.8)",
bordercolor="rgba(255,255,255,0.3)",
borderwidth=1,
),
hovermode="closest",
plot_bgcolor="rgba(0,0,0,0)",
paper_bgcolor="rgba(0,0,0,0)",
font=dict(color="white"),
)
fig.update_xaxes(showgrid=True, gridcolor="rgba(128,128,128,0.2)")
fig.update_yaxes(showgrid=True, gridcolor="rgba(128,128,128,0.2)")
return fig
def _make_audio_filename(engine_name: str, band: str, ext: str) -> str:
"""Generate a unique bucket filename for an audio file."""
from datetime import datetime
ts = datetime.now().strftime("%Y%m%d_%H%M%S")
safe_engine = engine_name.replace(" ", "_").replace("(", "").replace(")", "")
safe_band = band.replace("-", "")
return f"{ts}_{safe_engine}_{safe_band}{ext}"
# ββ event handlers ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def on_row_select(evt: gr.SelectData) -> tuple:
"""
On row click: play audio and show metrics detail card.
Uses _session_audio_urls indexed by row β URL never shown in table.
Falls back to load_history URLs if session list is shorter (history mode).
"""
try:
row_idx = evt.index[0]
# get audio url
url = None
if row_idx < len(_session_audio_urls):
url = _session_audio_urls[row_idx]
# get result for detail card
result = None
if row_idx < len(_session_results):
result = _session_results[row_idx]
# build detail markdown
if result:
detail = (
f"**Engine:** {result['engine']} | "
f"**Band:** {result['band']} | "
f"**Voice:** {result.get('voice', 'β')}\n\n"
f"**UTMOS:** {format_utmos(result['utmos'])} | "
f"**WER:** {format_wer(result['wer'])} | "
f"**RTF:** {format_rtf(result['rtf'])} | "
f"**Latency:** {result['latency_s']}s | "
f"**Cost:** {format_cost(result['engine_cost_usd'], result['chirp_equiv_usd'], result['engine'])}\n\n"
f"**Text:** {result.get('input_text', 'β')}"
)
else:
detail = ""
if url and str(url).startswith("http"):
return gr.update(value=url, visible=True), gr.update(value=detail, visible=True)
return gr.update(visible=False), gr.update(value=detail, visible=bool(detail))
except Exception as e:
print(f"[Playback] Row select failed: {e}")
return gr.update(visible=False), gr.update(visible=False)
def on_engine_change(engine_name: str):
"""Show voice dropdown only for Kokoro."""
is_kokoro = engine_name == "Kokoro (tuned)"
return gr.update(visible=is_kokoro)
def on_band_change(band: str, engine_name: str):
"""Update voice dropdown to recommended voice when band changes (Kokoro only)."""
if engine_name != "Kokoro (tuned)":
return gr.update(visible=False, value=KOKORO_DEFAULT_VOICE)
recommended = KOKORO_BAND_VOICE.get(band, KOKORO_DEFAULT_VOICE)
return gr.update(visible=True, value=recommended)
def run_synthesis(engine_name: str, band: str, text: str, voice: str):
if not text.strip():
yield None, "β Please enter some text first.", build_comparison_table(_session_results), build_business_chart(_session_results)
return
engine = ENGINE_MAP.get(engine_name)
if engine is None:
yield None, f"β Engine '{engine_name}' not found.", build_comparison_table(_session_results), build_business_chart(_session_results)
return
with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
tmp_path = f.name.replace(".wav", "")
yield None, f"Synthesizing with {engine_name}...", build_comparison_table(_session_results), build_business_chart(_session_results)
try:
# pass voice override only for Kokoro
if engine_name == "Kokoro (tuned)":
synth_result = engine.synthesize(text, band, tmp_path, voice_override=voice)
else:
synth_result = engine.synthesize(text, band, tmp_path)
audio_path = synth_result["audio_path"]
except NotImplementedError as e:
yield None, f"β {e}", build_comparison_table(_session_results), build_business_chart(_session_results)
return
except Exception as e:
yield None, f"β Synthesis failed: {e}", build_comparison_table(_session_results), build_business_chart(_session_results)
return
yield audio_path, "Running evals (WER, UTMOS, RTF)...", build_comparison_table(_session_results), build_business_chart(_session_results)
try:
eval_result = evaluate(
reference_text=text,
audio_path=audio_path,
latency_seconds=synth_result["latency_seconds"],
engine=engine,
band=band,
synth_voice=synth_result.get("voice", "unknown"),
actual_cost_usd=synth_result.get("actual_cost_usd", None),
)
except Exception as e:
yield audio_path, f"β Eval failed: {e}", build_comparison_table(_session_results), build_business_chart(_session_results)
return
# upload audio to Supabase in background β non-blocking
audio_ext = Path(audio_path).suffix
bucket_filename = _make_audio_filename(engine_name, band, audio_ext)
def _on_upload(url):
if url:
eval_result["audio_url"] = url
print(f"[Storage] Uploaded: {url}")
# update the CSV row with the real audio URL
try:
import pandas as pd
if os.path.exists(_EVAL_LOG_PATH):
df = pd.read_csv(_EVAL_LOG_PATH, dtype={"audio_url": str})
if "audio_url" not in df.columns:
df["audio_url"] = ""
# match by timestamp + engine + band β unique enough
mask = (
(df["timestamp"] == eval_result["timestamp"]) &
(df["engine"] == eval_result["engine"]) &
(df["band"] == eval_result["band"])
)
df.loc[mask, "audio_url"] = url
df.to_csv(_EVAL_LOG_PATH, index=False)
# re-upload updated CSV to Supabase
from storage import upload_csv_background
upload_csv_background(_EVAL_LOG_PATH)
except Exception as e:
print(f"[Storage] CSV audio_url update failed: {e}")
else:
eval_result["audio_url"] = None
upload_audio_background(audio_path, bucket_filename, callback=_on_upload)
eval_result["audio_url"] = None # placeholder until upload completes
_session_results.append(eval_result)
_session_audio_urls.append(eval_result.get("audio_url") or "")
status = (
f"β Done β "
f"UTMOS: {format_utmos(eval_result['utmos'])} | "
f"WER: {format_wer(eval_result['wer'])} | "
f"RTF: {format_rtf(eval_result['rtf'])}"
)
yield audio_path, status, build_comparison_table(_session_results), build_business_chart(_session_results)
def clear_results():
_session_results.clear()
_session_audio_urls.clear()
return build_comparison_table(_session_results), build_business_chart(_session_results), "Results cleared."
def export_session():
if not _session_results:
return gr.update(visible=False), "β No session results to export."
df = pd.DataFrame(_session_results)
export_path = os.path.join(os.path.dirname(__file__), "session_export.csv")
df.to_csv(export_path, index=False, encoding="utf-8-sig")
return gr.update(value=export_path, visible=True), "β Session exported."
def export_all():
if not os.path.exists(_EVAL_LOG_PATH):
return gr.update(visible=False), "β No history log found."
try:
df = pd.read_csv(_EVAL_LOG_PATH, dtype={"audio_url": str})
export_path = os.path.join(os.path.dirname(__file__), "history_export.csv")
df.to_csv(export_path, index=False, encoding="utf-8-sig")
return gr.update(value=export_path, visible=True), "β Full history log ready to download."
except Exception as e:
return gr.update(visible=False), f"β Failed: {e}"
def load_history():
global _session_results, _session_audio_urls
# try Supabase first, fall back to local CSV
try:
from storage import download_csv
download_csv(_EVAL_LOG_PATH)
except Exception as e:
print(f"[Storage] Supabase download skipped, using local: {e}")
if not os.path.exists(_EVAL_LOG_PATH):
return build_comparison_table([]), build_business_chart([]), "β No history found."
try:
df = pd.read_csv(_EVAL_LOG_PATH, dtype={"audio_url": str})
if "audio_url" not in df.columns:
df["audio_url"] = ""
records = df.to_dict(orient="records")
# populate session state so row click works
_session_results = records
_session_audio_urls = [
str(r.get("audio_url", "")) if str(r.get("audio_url", "")) not in ("nan", "None", "") else ""
for r in records
]
return build_comparison_table(records), build_business_chart(records), f"β Loaded {len(records)} historical runs."
except Exception as e:
return build_comparison_table([]), build_business_chart([]), f"β Failed: {e}"
def refresh_table():
"""Rebuild comparison table from current session results β picks up audio URLs from completed uploads."""
return build_comparison_table(_session_results)
# ββ UI ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
def build_ui():
with gr.Blocks(title="Bantrly TTS Evaluation Framework") as demo:
gr.Markdown("""
# π Bantrly TTS Evaluation Framework
Compare TTS engines on coaching text across grade bands.
**Metrics:** UTMOS (naturalness, β better) Β· WER (intelligibility, β better) Β· RTF (speed, β better) Β· Cost vs Chirp 3 HD
""")
with gr.Row():
with gr.Column(scale=1):
engine_selector = gr.Dropdown(
choices=ENGINE_CHOICES,
value=ENGINE_CHOICES[0],
label="TTS Engine",
)
band_selector = gr.Dropdown(
choices=BANDS,
value="K-2",
label="Grade Band",
)
voice_selector = gr.Dropdown(
choices=KOKORO_VOICES,
value=KOKORO_DEFAULT_VOICE,
label="Voice (Kokoro only)",
visible=True, # Kokoro is default engine
info="Defaults to recommended voice for selected band. Override freely.",
)
text_input = gr.Textbox(
label="Coaching Text",
placeholder="Type or paste any coaching text here...",
lines=4,
value="You did such a great job speaking today! I loved how loud and clear your voice was.",
)
synthesize_btn = gr.Button("βΆ Synthesize + Eval", variant="primary")
with gr.Column(scale=1):
audio_output = gr.Audio(label="Output Audio", type="filepath")
status_output = gr.Textbox(label="Status", interactive=False, lines=3)
gr.Markdown("## Comparison Table")
gr.Markdown(
"**β higher is better Β· β lower is better** β "
"WER may exceed 100% on short texts."
)
comparison_table = gr.Dataframe(
value=build_comparison_table([]),
label="Eval Results β click a row to play audio",
interactive=False,
)
with gr.Row():
with gr.Column(scale=1):
row_audio_player = gr.Audio(
label="βΆ Selected Row Audio",
visible=False,
type="filepath",
)
with gr.Column(scale=2):
row_detail = gr.Markdown(
value="",
visible=False,
)
business_chart = gr.Plot(
value=build_business_chart([]),
label="Business Decision Chart",
)
with gr.Row():
clear_btn = gr.Button("π Clear Session")
refresh_btn = gr.Button("π Refresh Table")
load_history_btn = gr.Button("π Load History")
export_session_btn = gr.Button("β¬ Export Session")
export_all_btn = gr.Button("β¬ Export Full History")
with gr.Row():
export_file = gr.File(label="Download CSV", visible=False)
export_status = gr.Textbox(label="", interactive=False, visible=True, value="")
# ββ bindings ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
engine_selector.change(
fn=on_engine_change,
inputs=[engine_selector],
outputs=[voice_selector],
)
band_selector.change(
fn=on_band_change,
inputs=[band_selector, engine_selector],
outputs=[voice_selector],
)
synthesize_btn.click(
fn=run_synthesis,
inputs=[engine_selector, band_selector, text_input, voice_selector],
outputs=[audio_output, status_output, comparison_table, business_chart],
)
clear_btn.click(
fn=clear_results,
outputs=[comparison_table, business_chart, export_status],
)
refresh_btn.click(
fn=refresh_table,
outputs=[comparison_table],
)
comparison_table.select(
fn=on_row_select,
inputs=[],
outputs=[row_audio_player, row_detail],
)
load_history_btn.click(
fn=load_history,
outputs=[comparison_table, business_chart, export_status],
)
export_session_btn.click(
fn=export_session,
outputs=[export_file, export_status],
)
export_all_btn.click(
fn=export_all,
outputs=[export_file, export_status],
)
return demo
if __name__ == "__main__":
demo = build_ui()
demo.launch(share=False) |