renderfy commited on
Commit
1c2f173
·
verified ·
1 Parent(s): bc1e133

Upload app.py

Browse files
Files changed (1) hide show
  1. app.py +57 -92
app.py CHANGED
@@ -1,6 +1,5 @@
1
- # app.py — Diag Studio AI Frontend (TR/EN, retries, warm-up, artifacts, Strict/Full, anomaly gate, adjudication)
2
  import os
3
- import io
4
  import json
5
  import time
6
  import requests
@@ -38,7 +37,7 @@ HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"} if HF_TOKEN else {}
38
  L = {
39
  "en": {
40
  "title": "Diag Studio AI — Engine Sound Diagnosis",
41
- "about": "Upload an engine sound (mp3/wav/ogg) or paste an audio URL. Backend runs fal.ai audio-understanding with detailed analysis, then summarizes with GPT-4o.",
42
  "backend_url": "Backend URL",
43
  "lang": "Language",
44
  "upload": "Upload audio file",
@@ -49,8 +48,6 @@ L = {
49
  "detailed": "Detailed analysis (always on)",
50
  "use_data_uri": "Send as data:URI (safer for private backends)",
51
  "model": "Commentary model (OpenAI)",
52
- "mode": "Prompt Mode",
53
- "mode_help": "Strict groups similar timbres to reduce mislabels; Full uses detailed taxonomy.",
54
  "analyze": "Analyze",
55
  "reset": "Reset",
56
  "result": "Result",
@@ -73,14 +70,14 @@ L = {
73
  "lang_help": "Output language for GPT-4o commentary",
74
  "ok_false": "Request returned a controlled error from backend.",
75
  "try_again": "Backend might be waking up or rate-limited. Try Analyze again or increase ‘Max total wait’.",
76
- "gate": "Anomaly Gate",
77
- "taxonomy": "Taxonomy",
78
- "applied_rules": "Adjudication rules applied",
79
- "confusions": "Common confusion pairs",
80
  },
81
  "tr": {
82
  "title": "Diag Studio AI — Motor Sesi Teşhis",
83
- "about": "Motor sesini (mp3/wav/ogg) yükleyin veya bir ses URL’si girin. Backend fal.ai ‘audio-understanding’i ayrıntılı analizle çalıştırır ve GPT-4o ile özetler.",
84
  "backend_url": "Backend URL",
85
  "lang": "Dil",
86
  "upload": "Ses dosyası yükle",
@@ -91,8 +88,6 @@ L = {
91
  "detailed": "Ayrıntılı analiz (her zaman açık)",
92
  "use_data_uri": "data:URI olarak gönder (özel backend için güvenli)",
93
  "model": "Yorum modeli (OpenAI)",
94
- "mode": "Prompt Modu",
95
- "mode_help": "Strict benzer tınıları gruplayıp karışmayı azaltır; Full detaylı taksonomi kullanır.",
96
  "analyze": "Analiz Et",
97
  "reset": "Sıfırla",
98
  "result": "Sonuç",
@@ -105,7 +100,7 @@ L = {
105
  "no_backend": "Bir backend URL girin (örn: http://127.0.0.1:7860)",
106
  "health_ok": "Backend erişilebilir.",
107
  "health_fail": "Backend’e erişilemedi.",
108
- "upload_hint": "Streamlit yükleme sınırı ~200 MB; mp3/wav/ogg kullanın.",
109
  "footer": "İpucu: Dosya yüklemek istemezseniz herkese açık bir ses URL’si de girebilirsiniz.",
110
  "auto_retry": "429/500/502/503/504 için otomatik yeniden dene",
111
  "warmup": "Analizden önce backend’i ısıt",
@@ -115,10 +110,10 @@ L = {
115
  "lang_help": "GPT-4o yorum çıktısının dili",
116
  "ok_false": "Backend kontrollü bir hata döndürdü.",
117
  "try_again": "Backend uyanıyor veya sınırlandı. Yeniden Dene’ye basın ya da ‘Azami bekleme’ değerini artırın.",
118
- "gate": "Arıza Kapısı (Anomaly Gate)",
119
- "taxonomy": "Taksonomi",
120
- "applied_rules": "Uygulanan adjudication kuralları",
121
- "confusions": "Sık karışan çiftler",
122
  }
123
  }
124
 
@@ -127,8 +122,6 @@ if "lang" not in st.session_state:
127
  st.session_state.lang = "tr"
128
  if "last_response" not in st.session_state:
129
  st.session_state.last_response = None
130
- if "audio_mem" not in st.session_state:
131
- st.session_state.audio_mem = None # keep last uploaded audio bytes for player
132
 
133
  # ================= Header =================
134
  col1, col2 = st.columns([0.6, 0.4])
@@ -150,12 +143,10 @@ api_base = st.text_input(t["backend_url"], value=backend_default)
150
 
151
  # ================= Advanced options =================
152
  with st.expander(t["adv"], expanded=False):
153
- st.checkbox(t["detailed"], value=True, disabled=True)
154
  use_data_uri = st.checkbox(t["use_data_uri"], value=True)
155
  openai_model = st.text_input(t["model"], value="gpt-4o")
156
  prompt_override = st.text_area(t["prompt"], value="", height=120)
157
- # NEW: Full/Strict selector
158
- mode = st.selectbox(t["mode"], ["FULL", "STRICT"], index=0, help=t["mode_help"])
159
  auto_retry = st.checkbox(t["auto_retry"], value=True)
160
  warmup = st.checkbox(t["warmup"], value=True)
161
  max_wait = st.slider(t["max_wait"], 10, 180, 60, step=10)
@@ -166,21 +157,15 @@ st.markdown("---")
166
  # ================= Upload / URL =================
167
  st.write(t["upload_hint"])
168
  file = st.file_uploader(t["upload"], type=["mp3", "wav", "ogg"])
169
- st.write(f"**{t['or']}**")
170
- audio_url = st.text_input(t["paste_url"], value="", placeholder="https://...")
171
-
172
- # If a file was uploaded, stash bytes for inline audio preview
173
  if file is not None:
174
  try:
175
- b = file.read()
176
- st.session_state.audio_mem = (b, file.type or "audio/mpeg")
177
- # Reset file pointer for requests
178
- file.seek(0)
179
  except Exception:
180
- st.session_state.audio_mem = None
181
 
182
- if st.session_state.audio_mem:
183
- st.audio(st.session_state.audio_mem[0], format=st.session_state.audio_mem[1])
184
 
185
  # ================= Controls =================
186
  c1, c2, c3 = st.columns([0.2, 0.2, 0.6])
@@ -227,14 +212,13 @@ def fetch_bytes(sess: requests.Session, base: str, url: str) -> bytes:
227
  r.raise_for_status()
228
  return r.content
229
 
230
- def post_analyze(sess: requests.Session, base: str, _file, _audio_url: str, lang: str, mode: str) -> dict:
231
  url = f"{base.rstrip('/')}/analyze"
232
  data = {
233
- "detailed_analysis": "true", # backend yine de True’ya sabitliyor
234
  "use_data_uri": "true" if use_data_uri else "false",
235
  "openai_model": openai_model or "gpt-4o",
236
- "lang": lang, # "tr" | "en"
237
- "prompt_mode": mode, # "FULL" | "STRICT"
238
  }
239
  if prompt_override.strip():
240
  data["prompt"] = prompt_override.strip()
@@ -295,13 +279,7 @@ if go:
295
  if warmup:
296
  warm_up_backend(sess, api_base, total_budget=min(max_wait, 40))
297
  with st.spinner(t["analyzing"]):
298
- # streamlit's UploadedFile can be used directly in requests; make sure pointer is at 0
299
- if file is not None:
300
- try:
301
- file.seek(0)
302
- except Exception:
303
- pass
304
- res = post_analyze(sess, api_base, file, audio_url, st.session_state.lang, mode)
305
  st.session_state.last_response = res
306
 
307
  # ================= Results =================
@@ -311,71 +289,58 @@ if res:
311
  st.error(t["ok_false"])
312
  st.write(res.get("error"))
313
  st.info(t["try_again"])
314
- # Debug blokları varsa göster
315
  if res.get("fal_debug"):
316
  st.expander("Debug").code(json.dumps(res["fal_debug"], ensure_ascii=False, indent=2), language="json")
317
  else:
318
  st.subheader(t["result"])
319
-
320
- # Anomaly Gate & Taxonomy quick view
321
- parsed = res.get("fal_parsed") or {}
322
- if isinstance(parsed, dict):
323
- gate = parsed.get("anomaly_gate") or {}
324
- colg1, colg2, colg3 = st.columns([0.33, 0.33, 0.34])
325
- with colg1:
326
- st.markdown(f"**{t['gate']}**")
327
- is_anom = gate.get("is_anomaly")
328
- score = gate.get("score")
329
- ev = gate.get("evidence")
330
- st.write(f"• is_anomaly: `{is_anom}`")
331
- st.write(f"• score: `{score}`")
332
- if ev: st.write(f"• evidence: `{ev}`")
333
- with colg2:
334
- st.markdown(f"**{t['taxonomy']}**")
335
- tax = parsed.get("taxonomy_version") or res.get("taxonomy_version")
336
- st.write(f"`{tax or '—'}`")
337
- pmode = res.get("prompt_mode")
338
- if pmode:
339
- st.write(f"mode: `{pmode}`")
340
- with colg3:
341
- rules = (parsed.get("adjudication") or {}).get("applied_rules") or (res.get("adjudication") or {}).get("applied_rules")
342
- st.markdown(f"**{t['applied_rules']}**")
343
- if rules:
344
- st.code(json.dumps(rules, ensure_ascii=False, indent=2), language="json")
345
- conf_pairs = (parsed.get("notes") or {}).get("confusion_pairs")
346
- if conf_pairs:
347
- st.markdown(f"**{t['confusions']}**")
348
- st.code(json.dumps(conf_pairs, ensure_ascii=False, indent=2), language="json")
349
-
350
- # Driver-friendly commentary
351
  comm = res.get("commentary") or ""
352
  if comm:
353
  st.markdown(f"**{t['commentary']}**")
354
  st.write(comm)
355
 
356
- # Raw parsed JSON
 
357
  if parsed:
358
  st.markdown(f"**{t['json']}**")
359
  st.json(parsed)
360
 
361
- # Backend artifacts
 
 
 
 
 
 
 
 
 
 
 
362
  st.markdown(f"**{t['backend_artifacts']}**")
363
  dls = res.get("downloads") or {}
364
  col_a, col_b, col_c = st.columns(3)
365
 
366
- def fetch_and_button(label: str, key: str, file_name: str, column):
367
- if dls.get(key):
368
- try:
369
- data = fetch_bytes(sess, api_base, dls[key])
370
- column.download_button(label=label, data=data, file_name=file_name)
371
- except Exception:
372
- column.error(f"{file_name} ✖")
373
- else:
374
- column.write("—")
 
 
 
 
375
 
376
- fetch_and_button(t["download_summary"], "summary_txt", "summary.txt", col_a)
377
- fetch_and_button(t["download_parsed"], "fal_parsed_json", "parsed.json", col_b)
378
- fetch_and_button(t["download_raw"], "fal_raw_txt", "raw.txt", col_c)
 
 
 
379
 
380
  # ===== Footer =====
381
- st.caption("© 2025 Diag Studio AI — Streamlit frontend (TR/EN, retries, warm-up, Strict/Full, anomaly gate, adjudication)")
 
1
+ # app.py — Diag Studio AI Frontend (TR/EN, retries, warm-up, artifacts download, audio preview)
2
  import os
 
3
  import json
4
  import time
5
  import requests
 
37
  L = {
38
  "en": {
39
  "title": "Diag Studio AI — Engine Sound Diagnosis",
40
+ "about": "Upload an engine sound (mp3/wav/ogg) or paste an audio URL. Backend runs fal.ai audio-understanding with detailed analysis, then maps to labels and summarizes with GPT-4o.",
41
  "backend_url": "Backend URL",
42
  "lang": "Language",
43
  "upload": "Upload audio file",
 
48
  "detailed": "Detailed analysis (always on)",
49
  "use_data_uri": "Send as data:URI (safer for private backends)",
50
  "model": "Commentary model (OpenAI)",
 
 
51
  "analyze": "Analyze",
52
  "reset": "Reset",
53
  "result": "Result",
 
70
  "lang_help": "Output language for GPT-4o commentary",
71
  "ok_false": "Request returned a controlled error from backend.",
72
  "try_again": "Backend might be waking up or rate-limited. Try Analyze again or increase ‘Max total wait’.",
73
+ "preview": "Audio preview",
74
+ "raw_text": "Raw (model text)",
75
+ "prompt_used": "Prompt used (first 240 chars)",
76
+ "api_used": "Backend audio URL (truncated)"
77
  },
78
  "tr": {
79
  "title": "Diag Studio AI — Motor Sesi Teşhis",
80
+ "about": "Motor sesini (mp3/wav/ogg) yükleyin veya bir ses URL’si girin. Backend fal.ai ‘audio-understanding’ ile ayrıntılı analiz yapar, etiketlere map eder ve GPT-4o ile özetler.",
81
  "backend_url": "Backend URL",
82
  "lang": "Dil",
83
  "upload": "Ses dosyası yükle",
 
88
  "detailed": "Ayrıntılı analiz (her zaman açık)",
89
  "use_data_uri": "data:URI olarak gönder (özel backend için güvenli)",
90
  "model": "Yorum modeli (OpenAI)",
 
 
91
  "analyze": "Analiz Et",
92
  "reset": "Sıfırla",
93
  "result": "Sonuç",
 
100
  "no_backend": "Bir backend URL girin (örn: http://127.0.0.1:7860)",
101
  "health_ok": "Backend erişilebilir.",
102
  "health_fail": "Backend’e erişilemedi.",
103
+ "upload_hint": "Streamlit varsayılan yükleme sınırı ~200 MB; mp3/wav/ogg kullanın.",
104
  "footer": "İpucu: Dosya yüklemek istemezseniz herkese açık bir ses URL’si de girebilirsiniz.",
105
  "auto_retry": "429/500/502/503/504 için otomatik yeniden dene",
106
  "warmup": "Analizden önce backend’i ısıt",
 
110
  "lang_help": "GPT-4o yorum çıktısının dili",
111
  "ok_false": "Backend kontrollü bir hata döndürdü.",
112
  "try_again": "Backend uyanıyor veya sınırlandı. Yeniden Dene’ye basın ya da ‘Azami bekleme’ değerini artırın.",
113
+ "preview": "Ses önizleme",
114
+ "raw_text": "Ham çıktı (model metni)",
115
+ "prompt_used": "Kullanılan prompt (ilk 240 karakter)",
116
+ "api_used": "Backend ses URL’si (kısaltılmış)"
117
  }
118
  }
119
 
 
122
  st.session_state.lang = "tr"
123
  if "last_response" not in st.session_state:
124
  st.session_state.last_response = None
 
 
125
 
126
  # ================= Header =================
127
  col1, col2 = st.columns([0.6, 0.4])
 
143
 
144
  # ================= Advanced options =================
145
  with st.expander(t["adv"], expanded=False):
146
+ st.checkbox(t["detailed"], value=True, disabled=True) # backend already enforces True
147
  use_data_uri = st.checkbox(t["use_data_uri"], value=True)
148
  openai_model = st.text_input(t["model"], value="gpt-4o")
149
  prompt_override = st.text_area(t["prompt"], value="", height=120)
 
 
150
  auto_retry = st.checkbox(t["auto_retry"], value=True)
151
  warmup = st.checkbox(t["warmup"], value=True)
152
  max_wait = st.slider(t["max_wait"], 10, 180, 60, step=10)
 
157
  # ================= Upload / URL =================
158
  st.write(t["upload_hint"])
159
  file = st.file_uploader(t["upload"], type=["mp3", "wav", "ogg"])
 
 
 
 
160
  if file is not None:
161
  try:
162
+ st.markdown(f"**{t['preview']}**")
163
+ st.audio(file, format=file.type or "audio/mpeg")
 
 
164
  except Exception:
165
+ pass
166
 
167
+ st.write(f"**{t['or']}**")
168
+ audio_url = st.text_input(t["paste_url"], value="", placeholder="https://...")
169
 
170
  # ================= Controls =================
171
  c1, c2, c3 = st.columns([0.2, 0.2, 0.6])
 
212
  r.raise_for_status()
213
  return r.content
214
 
215
+ def post_analyze(sess: requests.Session, base: str, _file, _audio_url: str, lang: str) -> dict:
216
  url = f"{base.rstrip('/')}/analyze"
217
  data = {
218
+ "detailed_analysis": "true", # backend pins True anyway
219
  "use_data_uri": "true" if use_data_uri else "false",
220
  "openai_model": openai_model or "gpt-4o",
221
+ "lang": lang, # pass language to backend (tr/en)
 
222
  }
223
  if prompt_override.strip():
224
  data["prompt"] = prompt_override.strip()
 
279
  if warmup:
280
  warm_up_backend(sess, api_base, total_budget=min(max_wait, 40))
281
  with st.spinner(t["analyzing"]):
282
+ res = post_analyze(sess, api_base, file, audio_url, st.session_state.lang)
 
 
 
 
 
 
283
  st.session_state.last_response = res
284
 
285
  # ================= Results =================
 
289
  st.error(t["ok_false"])
290
  st.write(res.get("error"))
291
  st.info(t["try_again"])
 
292
  if res.get("fal_debug"):
293
  st.expander("Debug").code(json.dumps(res["fal_debug"], ensure_ascii=False, indent=2), language="json")
294
  else:
295
  st.subheader(t["result"])
296
+ # driver-friendly commentary
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
297
  comm = res.get("commentary") or ""
298
  if comm:
299
  st.markdown(f"**{t['commentary']}**")
300
  st.write(comm)
301
 
302
+ # parsed JSON
303
+ parsed = res.get("fal_parsed")
304
  if parsed:
305
  st.markdown(f"**{t['json']}**")
306
  st.json(parsed)
307
 
308
+ # small extras: prompt used and fal audio url if present
309
+ extras_col1, extras_col2 = st.columns(2)
310
+ if res.get("prompt_used"):
311
+ with extras_col1:
312
+ st.caption(t["prompt_used"])
313
+ st.code(res["prompt_used"], language="text")
314
+ if res.get("fal_audio_url"):
315
+ with extras_col2:
316
+ st.caption(t["api_used"])
317
+ st.code(res["fal_audio_url"], language="text")
318
+
319
+ # backend artifacts
320
  st.markdown(f"**{t['backend_artifacts']}**")
321
  dls = res.get("downloads") or {}
322
  col_a, col_b, col_c = st.columns(3)
323
 
324
+ if dls.get("summary_txt"):
325
+ try:
326
+ data = fetch_bytes(sess, api_base, dls["summary_txt"])
327
+ col_a.download_button(label=t["download_summary"], data=data, file_name="summary.txt")
328
+ except Exception:
329
+ col_a.error("summary.txt ✖")
330
+
331
+ if dls.get("fal_parsed_json"):
332
+ try:
333
+ data = fetch_bytes(sess, api_base, dls["fal_parsed_json"])
334
+ col_b.download_button(label=t["download_parsed"], data=data, file_name="parsed.json", mime="application/json")
335
+ except Exception:
336
+ col_b.error("parsed.json ✖")
337
 
338
+ if dls.get("fal_raw_txt"):
339
+ try:
340
+ data = fetch_bytes(sess, api_base, dls["fal_raw_txt"])
341
+ col_c.download_button(label=t["download_raw"], data=data, file_name="raw.txt")
342
+ except Exception:
343
+ col_c.error("raw.txt ✖")
344
 
345
  # ===== Footer =====
346
+ st.caption("© 2025 Diag Studio AI — Streamlit frontend (TR/EN, retries, warm-up)")