when ref_text empty automatic transcribing

lpscr · Oct 30, 2024 · 3dd59b8 · 3dd59b8
1 parent 02d5913
commit 3dd59b8
Showing 1 changed file with 9 additions and 4 deletions.
diff --git a/src/f5_tts/api.py b/src/f5_tts/api.py
@@ -15,6 +15,9 @@
     infer_process,
     remove_silence_for_generated_wav,
     save_spectrogram,
+    preprocess_ref_audio_text,
+    target_sample_rate,
+    hop_length,
 )
 
 
@@ -31,10 +34,8 @@ def __init__(
     ):
         # Initialize parameters
         self.final_wave = None
-        self.target_sample_rate = 24000
-        self.n_mel_channels = 100
-        self.hop_length = 256
-        self.target_rms = 0.1
+        self.target_sample_rate = target_sample_rate
+        self.hop_length = hop_length
         self.seed = -1
 
         # Set device
@@ -97,6 +98,10 @@ def infer(
             seed = random.randint(0, sys.maxsize)
         seed_everything(seed)
         self.seed = seed
+
+        if ref_text == "":
+            ref_file, ref_text = preprocess_ref_audio_text(ref_file, ref_text, device=self.device)
+
         wav, sr, spect = infer_process(
             ref_file,
             ref_text,