bluenevus's picture
Update app.py
0f94ce5 verified
import dash
from dash import dcc, html, Input, Output, State, callback
import dash_bootstrap_components as dbc
import base64
import io
import os
from snac import SNAC
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import google.generativeai as genai
import re
import logging
import numpy as np
from pydub import AudioSegment
from docx import Document
import PyPDF2
from tqdm import tqdm
import soundfile as sf
# Initialize logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
# Initialize device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Load models
print("Loading SNAC model...")
snac_model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz")
snac_model = snac_model.to(device)
model_name = "canopylabs/orpheus-3b-0.1-ft"
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
model.to(device)
tokenizer = AutoTokenizer.from_pretrained(model_name)
print(f"Orpheus model loaded to {device}")
# Available voices and emotive tags
VOICES = ["tara", "leah", "jess", "leo", "dan", "mia", "zac", "zoe"]
EMOTIVE_TAGS = ["<laugh>", "<chuckle>", "<sigh>", "<cough>", "<sniffle>", "<groan>", "<yawn>", "<gasp>"]
# Initialize Dash app
app = dash.Dash(__name__, external_stylesheets=[dbc.themes.BOOTSTRAP])
app.layout = dbc.Container([
dbc.Row([
dbc.Col([
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"),
dbc.Input(id="podcast-topic", placeholder="Enter podcast topic", className="mb-2"),
dbc.Textarea(id="prompt", placeholder="Enter your text here...", rows=5, className="mb-2"),
dcc.Upload(
id='upload-file',
children=html.Div(['Drag and Drop or ', html.A('Select a File')]),
style={
'width': '100%',
'height': '60px',
'lineHeight': '60px',
'borderWidth': '1px',
'borderStyle': 'dashed',
'borderRadius': '5px',
'textAlign': 'center',
'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"]],
value="1",
inline=True,
className="mb-2"
),
dbc.Button("Generate Podcast Script", id="generate-script-btn", color="primary", className="mb-2"),
dbc.Spinner(html.Div(id="script-loading"), color="primary"),
], 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 d-block"),
html.Label("Voice 1", className="mt-3"),
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"),
dbc.Spinner(html.Div(id="audio-loading"), color="primary"),
html.Div(id="audio-output"),
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),
], width=6),
]),
dcc.Store(id='generated-script'),
dcc.Store(id='generated-audio'),
])
def process_prompt(prompt, voice, tokenizer, device):
prompt = f"{voice}: {prompt}"
input_ids = tokenizer(prompt, return_tensors="pt").input_ids
start_token = torch.tensor([[128259]], dtype=torch.int64)
end_tokens = torch.tensor([[128009, 128260]], dtype=torch.int64)
modified_input_ids = torch.cat([start_token, input_ids, end_tokens], dim=1)
attention_mask = torch.ones_like(modified_input_ids)
return modified_input_ids.to(device), attention_mask.to(device)
def parse_output(generated_ids):
token_to_find = 128257
token_to_remove = 128258
token_indices = (generated_ids == token_to_find).nonzero(as_tuple=True)
if len(token_indices[1]) > 0:
last_occurrence_idx = token_indices[1][-1].item()
cropped_tensor = generated_ids[:, last_occurrence_idx+1:]
else:
cropped_tensor = generated_ids
processed_rows = []
for row in cropped_tensor:
masked_row = row[row != token_to_remove]
processed_rows.append(masked_row)
code_lists = []
for row in processed_rows:
row_length = row.size(0)
new_length = (row_length // 7) * 7
trimmed_row = row[:new_length]
trimmed_row = [t - 128266 for t in trimmed_row]
code_lists.append(trimmed_row)
return code_lists[0]
def redistribute_codes(code_list, snac_model):
device = next(snac_model.parameters()).device # Get the device of SNAC model
layer_1 = []
layer_2 = []
layer_3 = []
for i in range((len(code_list)+1)//7):
layer_1.append(code_list[7*i])
layer_2.append(code_list[7*i+1]-4096)
layer_3.append(code_list[7*i+2]-(2*4096))
layer_3.append(code_list[7*i+3]-(3*4096))
layer_2.append(code_list[7*i+4]-(4*4096))
layer_3.append(code_list[7*i+5]-(5*4096))
layer_3.append(code_list[7*i+6]-(6*4096))
codes = [
torch.tensor(layer_1, device=device).unsqueeze(0),
torch.tensor(layer_2, device=device).unsqueeze(0),
torch.tensor(layer_3, device=device).unsqueeze(0)
]
audio_hat = snac_model.decode(codes)
return audio_hat.detach().squeeze().cpu().numpy() # Always return CPU numpy array
def detect_silence(audio, threshold=0.005, min_silence_duration=1.3):
sample_rate = 24000 # Adjust if your sample rate is different
is_silent = np.abs(audio) < threshold
silent_regions = np.where(is_silent)[0]
silence_starts = []
silence_ends = []
if len(silent_regions) > 0:
silence_starts.append(silent_regions[0])
for i in range(1, len(silent_regions)):
if silent_regions[i] - silent_regions[i-1] > 1:
silence_ends.append(silent_regions[i-1])
silence_starts.append(silent_regions[i])
silence_ends.append(silent_regions[-1])
long_silences = [(start, end) for start, end in zip(silence_starts, silence_ends)
if (end - start) / sample_rate >= min_silence_duration]
return long_silences
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 tqdm(enumerate(paragraphs), total=len(paragraphs), desc="Generating audio"):
if not paragraph.strip():
continue
voice = voice1 if num_hosts == "1" or i % 2 == 0 else voice2
input_ids, attention_mask = process_prompt(paragraph, voice, tokenizer, device)
with torch.no_grad():
generated_ids = model.generate(
input_ids,
attention_mask=attention_mask,
do_sample=True,
temperature=temperature,
top_p=top_p,
repetition_penalty=repetition_penalty,
max_new_tokens=max_new_tokens,
num_return_sequences=1,
eos_token_id=128258,
)
code_list = parse_output(generated_ids)
paragraph_audio = redistribute_codes(code_list, snac_model)
# Add silence detection here
silences = detect_silence(paragraph_audio)
if silences:
# Trim the audio at the last detected silence
paragraph_audio = paragraph_audio[:silences[-1][1]]
audio_samples.append(paragraph_audio)
final_audio = np.concatenate(audio_samples)
# Normalize the audio
final_audio = np.int16(final_audio / np.max(np.abs(final_audio)) * 32767)
return final_audio
except Exception as e:
logger.error(f"Error generating speech: {str(e)}")
return None
@callback(
Output("script-output", "value"),
Output("audio-output", "children"),
Output("advanced-settings", "is_open"),
Output("prompt", "value"),
Output("script-loading", "children"),
Output("audio-loading", "children"),
Input("generate-script-btn", "n_clicks"),
Input("generate-audio-btn", "n_clicks"),
Input("advanced-settings-toggle", "n_clicks"),
Input("clear-btn", "n_clicks"),
State("host1-name", "value"),
State("host2-name", "value"),
State("podcast-name", "value"),
State("podcast-topic", "value"),
State("prompt", "value"),
State("upload-file", "contents"),
State("duration", "value"),
State("num-hosts", "value"),
State("script-output", "value"),
State("voice1", "value"),
State("voice2", "value"),
State("temperature", "value"),
State("top-p", "value"),
State("repetition-penalty", "value"),
State("max-new-tokens", "value"),
State("advanced-settings", "is_open"),
prevent_initial_call=True
)
def combined_callback(generate_script_clicks, generate_audio_clicks, advanced_settings_clicks, clear_clicks,
host1_name, host2_name, podcast_name, podcast_topic, prompt, uploaded_file, duration, num_hosts,
script_output, voice1, voice2, temperature, top_p, repetition_penalty, max_new_tokens, is_advanced_open):
ctx = dash.callback_context
if not ctx.triggered:
return dash.no_update, dash.no_update, dash.no_update, dash.no_update, "", ""
trigger_id = ctx.triggered[0]['prop_id'].split('.')[0]
if trigger_id == "advanced-settings-toggle":
return dash.no_update, dash.no_update, not is_advanced_open, dash.no_update, "", ""
if trigger_id == "generate-script-btn":
try:
api_key = os.environ.get("GEMINI_API_KEY")
if not api_key:
raise ValueError("Gemini API key not found in environment variables")
genai.configure(api_key=api_key)
model = genai.GenerativeModel('gemini-2.5-pro-preview-03-25')
combined_content = prompt or ""
if uploaded_file:
content_type, content_string = uploaded_file.split(',')
decoded = base64.b64decode(content_string)
file_bytes = io.BytesIO(decoded)
file_bytes.seek(0)
if file_bytes.read(4) == b'%PDF':
file_bytes.seek(0)
pdf_reader = PyPDF2.PdfReader(file_bytes)
file_content = "\n".join([page.extract_text() for page in pdf_reader.pages])
else:
file_bytes.seek(0)
try:
file_content = file_bytes.read().decode('utf-8')
except UnicodeDecodeError:
file_bytes.seek(0)
try:
doc = Document(file_bytes)
file_content = "\n".join([para.text for para in doc.paragraphs])
except:
raise ValueError("Unsupported file type or corrupted file")
combined_content += "\n" + file_content if combined_content else file_content
num_hosts = int(num_hosts) if num_hosts else 1
prompt_template = f"""
Create a podcast script for {num_hosts} {'person' if num_hosts == 1 else 'people'} discussing:
{combined_content}
Duration: {duration} minutes. Include natural speech, humor, and occasional off-topic thoughts.
Use speech fillers like um, ah. Vary emotional tone.
Format: {'Monologue' if num_hosts == 1 else 'Alternating dialogue'} without speaker labels.
Separate {'paragraphs' if num_hosts == 1 else 'lines'} with blank lines.
If the number of {num_hosts} is 1 then each paragraph will be no more than 3 sentences each
Only provide the dialog for text to speech.
Only use these emotion tags in angle brackets: {', '.join(EMOTIVE_TAGS)}.
-Example: "I can't believe I stayed up all night <yawn> only to find out the meeting was canceled <groan>."
Ensure content flows naturally and stays on topic. Match the script length to {duration} minutes.
Do not include speaker labels like "jane:" or "john:" before dialogue.
The intro always includes the ({host1_name} and/or {host2_name}) if it exists and should be in the same paragraph.
The outro always includes the ({host1_name} and/or {host2_name}) if it exists and should be in the same paragraph
Do not include these types of transitions in the intro, outro or between paragraphs for example: "Intro Music fades in...". Its just dialog.
Keep each speaker's entire monologue in a single paragraph, regardless of length if the number of hosts is not 1.
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 WAV format
buffer = io.BytesIO()
sf.write(buffer, final_audio, 24000, format='WAV', subtype='PCM_16')
buffer.seek(0)
# Convert to base64 for audio playback
audio_base64 = base64.b64encode(buffer.getvalue()).decode('utf-8')
src = f"data:audio/wav;base64,{audio_base64}"
# Log audio file size
logger.info(f"Generated audio file size: {len(audio_base64)} bytes")
# 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:
logger.error("Failed to generate audio")
return dash.no_update, html.Div("Error generating audio"), dash.no_update, 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.")