Update and #26

app.py

@@ -29,9 +29,11 @@ import sys
 import torch
 import fairseq
 from picklescan.scanner import scan_file_path
+import hashlib
 
 
 ALLOWED_DOMAINS = {"huggingface.co", "hf.co"}
+MODEL_CACHE = {}
 
 parser = argparse.ArgumentParser(description="Run the app with optional sharing")
 parser.add_argument(
@@ -109,6 +111,14 @@ def check_model_safety(file_path: str):
             print(f"Integrity check skipped for {file_path}: {e}")
 
 
+def get_file_hash(file_path: str) -> str:
+    hasher = hashlib.sha256()
+    with open(file_path, "rb") as f:
+        while chunk := f.read(65536):
+            hasher.update(chunk)
+    return hasher.hexdigest()
+
+
 async def get_voices_list(proxy=None):
     """Print all available voices."""
     from edge_tts import list_voices
@@ -156,6 +166,26 @@ def find_my_model(a_, b_):
             check_model_safety(a_)
         return a_, b_
 
+    input_hash = None
+    if a_ and os.path.exists(a_):
+        input_hash = get_file_hash(a_)
+        if b_ and os.path.exists(b_):
+            input_hash += "_" + get_file_hash(b_)
+
+    if input_hash and input_hash in MODEL_CACHE:
+        cached_model, cached_index = MODEL_CACHE[input_hash]
+        model_exists = cached_model and os.path.exists(cached_model)
+        index_exists = (cached_index is None) or os.path.exists(cached_index)
+
+        if model_exists and index_exists:
+            check_model_safety(cached_model)
+            gr.Info(f"Model found: {cached_model}")
+            if cached_index:
+                gr.Info(f"Index found: {cached_index}")
+            return cached_model, cached_index
+        else:
+            del MODEL_CACHE[input_hash]
+
     txt_files = []
     for base_file in [a_, b_]:
         if base_file is not None and base_file.endswith(".txt"):
@@ -200,6 +230,9 @@ def find_my_model(a_, b_):
     if not index:
         gr.Warning("Index not found")
 
+    if model and input_hash:
+        MODEL_CACHE[input_hash] = (model, index)
+
     return model, index
 
 
@@ -471,7 +504,6 @@ def run(
     audio_effects,
     type_output,
     steps,
-    play_audio=False,
 ):
     if not audio_files:
         raise ValueError("The audio pls")
@@ -487,7 +519,7 @@ def run(
 
     if file_m is not None and (file_m.endswith(".txt") or file_m.endswith(".zip")):
         file_m, file_index = find_my_model(file_m, file_index)
-        print(file_m, file_index)
+        # print(file_m, file_index)
 
     random_tag = "USER_"+str(random.randint(10000000, 99999999))
 
@@ -513,9 +545,37 @@ def run(
     if audio_effects:
         result = add_audio_effects(result, type_output)
 
-    player_audio = result[0] if (play_audio and result and len(result) > 0) else None
+    return result
+
+
+def clear_player():
+    return None
+
+
+def load_first_audio(output_files, play_audio):
+    if not play_audio or not output_files:
+        return None
+    first_file = output_files[0]
+    if isinstance(first_file, dict):
+        first_file = first_file.get("name")
+    return first_file
+
+
+def audio_source_conf():
+    return gr.Radio(
+        choices=["📁 Upload", "🗣️ TTS", "🎙️ Record"],
+        value="📁 Upload",
+        label="Audio Source",
+    )
 
-    return result, player_audio
+
+def mic_conf():
+    return gr.Audio(
+        sources=["microphone"],
+        type="filepath",
+        label="Record Audio",
+        visible=False,
+    )
 
 
 def audio_conf():
@@ -623,15 +683,6 @@ def output_conf():
     )
 
 
-def active_tts_conf():
-    return gr.Checkbox(
-        False,
-        label="TTS",
-        # info="",
-        container=False,
-    )
-
-
 def tts_voice_conf():
     return gr.Dropdown(
         label="tts voice",
@@ -730,16 +781,21 @@ def infer_tts_audio(tts_voice, tts_text, play_audio):
     return [out_path], player_audio
 
 
-def show_components_tts(value_active):
-    return gr.update(
-        visible=value_active
-    ), gr.update(
-        visible=value_active
-    ), gr.update(
-        visible=value_active
+def show_components_source(mode):
+    return (
+        gr.update(visible=(mode == "🗣️ TTS")),
+        gr.update(visible=(mode == "🗣️ TTS")),
+        gr.update(visible=(mode == "🗣️ TTS")),
+        gr.update(visible=(mode == "🎙️ Record")),
     )
 
 
+def sync_mic_audio(mic_path):
+    if not mic_path:
+        return None
+    return [mic_path]
+
+
 def down_active_conf():
     return gr.Checkbox(
         False,
@@ -787,7 +843,7 @@ def get_gui(theme):
         gr.Markdown(title)
         gr.Markdown(description)
 
-        active_tts = active_tts_conf()
+        audio_source = audio_source_conf()
         with gr.Row():
             with gr.Column(scale=1):
                 tts_text = tts_text_conf()
@@ -799,15 +855,22 @@ def get_gui(theme):
 
                 tts_button = tts_button_conf()
 
-        active_tts.change(
-            fn=show_components_tts,
-            inputs=[active_tts],
-            outputs=[tts_voice, tts_text, tts_button],
-        )
-
+        mic_aud = mic_conf()
         aud = audio_conf()
         # gr.HTML("<hr>")
 
+        audio_source.change(
+            fn=show_components_source,
+            inputs=[audio_source],
+            outputs=[tts_voice, tts_text, tts_button, mic_aud],
+        )
+
+        mic_aud.change(
+            fn=sync_mic_audio,
+            inputs=[mic_aud],
+            outputs=[aud],
+        )
+
         down_active_gui = down_active_conf()
         down_info = gr.Markdown(
             f"Provide a link to a zip file, like this one: `https://huggingface.co/MrDawg/ToothBrushing/resolve/main/ToothBrushing.zip?download=true`, or separate links with a comma for the .pth and .index files, like this: `{test_model}`",
@@ -885,9 +948,17 @@ def get_gui(theme):
                 effects_gui,
                 format_out,
                 steps_gui,
-                player_gui,
             ],
-            outputs=[output_base, player_audio],
+            outputs=[output_base],
+        ).success(
+            clear_player,
+            inputs=None,
+            outputs=[player_audio],
+            queue=False,
+        ).success(
+            load_first_audio,
+            inputs=[output_base, player_gui],
+            outputs=[player_audio],
         )
 
         gr.Examples(
@@ -939,7 +1010,7 @@ def get_gui(theme):
                 envel_r,
                 const,
             ],
-            outputs=[output_base, player_audio],
+            outputs=[output_base],
             cache_examples=False,
         )
         gr.Markdown(RESOURCES)
@@ -948,12 +1019,16 @@ def get_gui(theme):
 
 
 if __name__ == "__main__":
-    tts_voice_list = asyncio.new_event_loop().run_until_complete(get_voices_list(proxy=None))
-    voices = sorted([
-        (" - ".join(reversed(v["FriendlyName"].split("-"))).replace("Microsoft ", "").replace("Online (Natural)", f"({v['Gender']})").strip(), f"{v['ShortName']}-{v['Gender']}")
-        for v in tts_voice_list
-    ])
-
+    try:
+        tts_voice_list = asyncio.new_event_loop().run_until_complete(get_voices_list(proxy=None))
+        voices = sorted([
+            (" - ".join(reversed(v["FriendlyName"].split("-"))).replace("Microsoft ", "").replace("Online (Natural)", f"({v['Gender']})").strip(), f"{v['ShortName']}-{v['Gender']}")
+            for v in tts_voice_list
+        ])
+    except Exception as e:
+        print(f"Warning: Could not retrieve online voices ({e}). Using default preset.")
+        voices = [("English (United States) - EmmaMultilingual (Female)", "en-US-EmmaMultilingualNeural-Female")]
+    
     app = get_gui(theme)
 
     app.queue(default_concurrency_limit=40)