Update app.py

app.py

@@ -14,7 +14,7 @@ import numpy as np
 from pydub import AudioSegment
 from docx import Document
 import PyPDF2
-import spaces
+from tqdm import tqdm
 
 # Initialize logging
 logging.basicConfig(level=logging.INFO)
@@ -45,7 +45,11 @@ app = dash.Dash(__name__, external_stylesheets=[dbc.themes.BOOTSTRAP])
 app.layout = dbc.Container([
     dbc.Row([
         dbc.Col([
-            html.H1("Orpheus Text-to-Speech", className="mb-4"),
+            html.H1("Orpheus Text-to-Speech", className="text-center mb-4"),
+        ], width=12),
+    ]),
+    dbc.Row([
+        dbc.Col([
             dbc.Input(id="host1-name", placeholder="Enter name of first host", className="mb-2"),
             dbc.Input(id="host2-name", placeholder="Enter name of second host", className="mb-2"),
             dbc.Input(id="podcast-name", placeholder="Enter podcast name", className="mb-2"),
@@ -65,7 +69,9 @@ app.layout = dbc.Container([
                     'margin': '10px 0'
                 },
             ),
+            html.Label("Duration (minutes)", className="mt-2"),
             dcc.Slider(id="duration", min=1, max=60, value=5, step=1, marks={1: '1', 30: '30', 60: '60'}, className="mb-2"),
+            html.Label("Number of Hosts", className="mt-2"),
             dbc.RadioItems(
                 id="num-hosts",
                 options=[{"label": i, "value": i} for i in ["1", "2"]],
@@ -77,18 +83,24 @@ app.layout = dbc.Container([
         ], width=6),
         dbc.Col([
             dbc.Textarea(id="script-output", placeholder="Generated script will appear here...", rows=10, className="mb-2"),
+            dbc.Button("Clear", id="clear-btn", color="secondary", className="mb-2"),
+            html.Label("Voice 1", className="mt-2"),
             dcc.Dropdown(id="voice1", options=[{"label": v, "value": v} for v in VOICES], value="tara", className="mb-2"),
+            html.Label("Voice 2", className="mt-2"),
             dcc.Dropdown(id="voice2", options=[{"label": v, "value": v} for v in VOICES], value="zac", className="mb-2"),
             dbc.Button("Generate Audio", id="generate-audio-btn", color="success", className="mb-2"),
             html.Div(id="audio-output"),
-            dbc.Button("Clear", id="clear-btn", color="secondary", className="mb-2"),
+            dbc.Button("Advanced Settings", id="advanced-settings-toggle", color="info", className="mb-2"),
             dbc.Collapse([
+                html.Label("Temperature", className="mt-2"),
                 dcc.Slider(id="temperature", min=0.1, max=1.5, value=0.6, step=0.05, marks={0.1: '0.1', 0.8: '0.8', 1.5: '1.5'}, className="mb-2"),
+                html.Label("Top P", className="mt-2"),
                 dcc.Slider(id="top-p", min=0.1, max=1.0, value=0.9, step=0.05, marks={0.1: '0.1', 0.5: '0.5', 1.0: '1.0'}, className="mb-2"),
+                html.Label("Repetition Penalty", className="mt-2"),
                 dcc.Slider(id="repetition-penalty", min=1.0, max=2.0, value=1.2, step=0.1, marks={1.0: '1.0', 1.5: '1.5', 2.0: '2.0'}, className="mb-2"),
+                html.Label("Max New Tokens", className="mt-2"),
                 dcc.Slider(id="max-new-tokens", min=100, max=16384, value=4096, step=100, marks={100: '100', 8192: '8192', 16384: '16384'}, className="mb-2"),
             ], id="advanced-settings", is_open=False),
-            dbc.Button("Advanced Settings", id="advanced-settings-toggle", color="info", className="mb-2"),
         ], width=6),
     ]),
     dcc.Store(id='generated-script'),
@@ -126,13 +138,12 @@ def detect_silence(audio, threshold=0.01, min_silence_len=1000):
         silent_regions.append((silent_start, len(audio)))
     return silent_regions
 
-@spaces.GPU()
 def generate_audio(script_output, voice1, voice2, num_hosts, temperature, top_p, repetition_penalty, max_new_tokens):
     try:
         paragraphs = script_output.split('\n\n')  # Split by double newline
         audio_samples = []
         
-        for i, paragraph in enumerate(paragraphs):
+        for i, paragraph in tqdm(enumerate(paragraphs), total=len(paragraphs), desc="Generating audio"):
             if not paragraph.strip():
                 continue
             
@@ -151,6 +162,7 @@ def generate_audio(script_output, voice1, voice2, num_hosts, temperature, top_p,
                     max_new_tokens=max_new_tokens,
                     num_return_sequences=1,
                     eos_token_id=128258,
+                    pad_token_id=128258,
                 )
             
             code_list = parse_output(generated_ids)
@@ -265,52 +277,4 @@ def combined_callback(generate_script_clicks, generate_audio_clicks, advanced_se
             Start a new paragraph only when switching to a different speaker if the number of hosts is not 1.
             Maintain natural conversation flow and speech patterns within each monologue.
             Use context clues or subtle references to indicate who is speaking without explicit labels if the number of hosts is not 1.
-            Use speaker names ({host1_name} and/or {host2_name}) sparingly, only when necessary for clarity or emphasis. Avoid starting every line with the other person's name.
-            Rely more on context and speech patterns to indicate who is speaking, rather than always stating names.
-            Use names primarily for transitions sparingly, definitely with agreements, or to draw attention to a specific point, not as a constant form of address.
-            {'Make sure the script is a monologue for one person.' if num_hosts == 1 else f'Ensure the dialogue alternates between two distinct voices, with {host1_name} speaking on odd-numbered lines and {host2_name} on even-numbered lines.'}
-            Always include intro with the speaker name and its the podcast name "{podcast_name}" in intoduce the topic of the podcast with "{podcast_topic}".
-            Incorporate the podcast name and topic naturally into the intro and outro, and ensure the content stays relevant to the specified topic throughout the script.
-            """
-            
-            response = model.generate_content(prompt_template)
-            return re.sub(r'[^a-zA-Z0-9\s.,?!<>]', '', response.text), dash.no_update, dash.no_update, dash.no_update
-        except Exception as e:
-            logger.error(f"Error generating podcast script: {str(e)}")
-            return f"Error: {str(e)}", dash.no_update, dash.no_update, dash.no_update
-
-    elif trigger_id == "generate-audio-btn":
-        if not script_output.strip():
-            return dash.no_update, html.Div("No audio generated yet."), dash.no_update, dash.no_update
-        
-        final_audio = generate_audio(script_output, voice1, voice2, num_hosts, temperature, top_p, repetition_penalty, max_new_tokens)
-        
-        if final_audio is not None:
-            # Convert to base64 for audio playback
-            audio_base64 = base64.b64encode(final_audio.tobytes()).decode('utf-8')
-            src = f"data:audio/wav;base64,{audio_base64}"
-            
-            # Create a download link for the audio
-            download_link = html.A("Download Audio", href=src, download="generated_audio.wav")
-            
-            return dash.no_update, html.Div([
-                html.Audio(src=src, controls=True),
-                html.Br(),
-                download_link
-            ]), dash.no_update, dash.no_update
-        else:
-            return dash.no_update, html.Div("Error generating audio"), dash.no_update, dash.no_update
-
-    elif trigger_id == "advanced-settings-toggle":
-        return dash.no_update, dash.no_update, not is_advanced_open, dash.no_update
-
-    elif trigger_id == "clear-btn":
-        return "", html.Div("No audio generated yet."), dash.no_update, ""
-
-    return dash.no_update, dash.no_update, dash.no_update, dash.no_update
-
-# Run the app
-if __name__ == '__main__':
-    print("Starting the Dash application...")
-    app.run(debug=True, host='0.0.0.0', port=7860)
-    print("Dash application has finished running.")
+            Use speaker names ({host1_name} and/or {host2_