update logs

app/server.py

@@ -1,4 +1,5 @@
 import base64
+import logging
 import os
 import re
 import subprocess
@@ -75,6 +76,7 @@ STATE: dict[str, Any] = {
 }
 
 ANSI_RE = re.compile(r"\x1B\[[0-9;]*[A-Za-z]")
+LOGGER = logging.getLogger("uvicorn.error")
 
 
 def read_vocab(path: str) -> dict[str, str]:
@@ -154,16 +156,21 @@ def decode_tokens(vocab: dict[str, str], token_ids: list[int], language: str) ->
     return text
 
 
-def transcribe_file(path: str, language: str) -> str:
+def transcribe_file(path: str, language: str) -> tuple[str, dict[str, float | int]]:
+    t0 = time.perf_counter()
     waveform, sr = sf.read(path)
     waveform = to_mono(np.asarray(waveform, dtype=np.float32))
     waveform = ensure_sample_rate(waveform, sr)
     mel = log_mel_spectrogram(waveform, STATE["mel_filters"])
     encoder_input = pad_or_trim(mel)
+    t_pre = time.perf_counter()
 
     encoded = STATE["encoder"].run(None, {"x": encoder_input})[0]
+    t_enc = time.perf_counter()
 
     tokens = [50258, TASK_CODE[language], 50359, 50363]
+    tokens = tokens * 3
+    pop_index = 12
     emitted: list[int] = []
 
     for _ in range(MAX_DECODE_TOKENS):
@@ -178,11 +185,26 @@ def transcribe_file(path: str, language: str) -> str:
         if next_token == END_TOKEN:
             break
         tokens.append(next_token)
+
+        if pop_index > 4:
+            pop_index -= 1
+        tokens.pop(pop_index)
+
         if next_token <= TIMESTAMP_BEGIN:
             emitted.append(next_token)
 
     vocab = STATE["vocab_en"] if language == "en" else STATE["vocab_zh"]
-    return decode_tokens(vocab, emitted, language)
+    text = decode_tokens(vocab, emitted, language)
+    t_end = time.perf_counter()
+    stats: dict[str, float | int] = {
+        "audio_sec": float(len(waveform) / SAMPLE_RATE),
+        "pre_ms": float((t_pre - t0) * 1000.0),
+        "enc_ms": float((t_enc - t_pre) * 1000.0),
+        "dec_ms": float((t_end - t_enc) * 1000.0),
+        "total_ms": float((t_end - t0) * 1000.0),
+        "tokens": int(len(emitted)),
+    }
+    return text, stats
 
 
 def convert_to_wav(src_path: str) -> str:
@@ -413,13 +435,28 @@ async def transcriptions(
         with open(input_path, "wb") as f:
             f.write(payload)
         wav_path = convert_to_wav(input_path)
-        text = transcribe_file(wav_path, language)
+        text, perf = transcribe_file(wav_path, language)
     finally:
         if os.path.exists(input_path):
             os.unlink(input_path)
         if wav_path and os.path.exists(wav_path):
             os.unlink(wav_path)
 
+    total_s = perf["total_ms"] / 1000.0 if perf["total_ms"] else 0.0
+    rtf = (total_s / perf["audio_sec"]) if perf["audio_sec"] else 0.0
+    LOGGER.info(
+        "stt_perf model=%s lang=%s audio_sec=%.3f pre_ms=%.1f enc_ms=%.1f dec_ms=%.1f total_ms=%.1f rtf=%.3f tokens=%d",
+        model,
+        language,
+        perf["audio_sec"],
+        perf["pre_ms"],
+        perf["enc_ms"],
+        perf["dec_ms"],
+        perf["total_ms"],
+        rtf,
+        perf["tokens"],
+    )
+
     if response_format == "text":
         return PlainTextResponse(text)
     if response_format == "verbose_json":