Spaces:
Running
Running
updates
Browse files
app.py
CHANGED
@@ -2,8 +2,8 @@ import streamlit as st
|
|
2 |
import numpy as np
|
3 |
import torch
|
4 |
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
|
5 |
-
import soundfile as sf
|
6 |
from io import StringIO
|
|
|
7 |
|
8 |
# Load models outside of function calls for efficiency
|
9 |
@st.cache_data
|
@@ -23,30 +23,31 @@ def get_speaker_embeddings():
|
|
23 |
|
24 |
speaker_embeddings = get_speaker_embeddings()
|
25 |
|
26 |
-
# Improved Styling
|
27 |
def local_css(file_name):
|
28 |
with open(file_name) as f:
|
29 |
st.markdown(f'<style>{f.read()}</style>', unsafe_allow_html=True)
|
30 |
|
31 |
-
local_css("style.css")
|
32 |
|
33 |
-
#
|
34 |
st.title("Text-to-Voice Conversion")
|
35 |
st.markdown("Convert your text to speech using advanced AI models.")
|
36 |
|
37 |
# Function to convert text to speech
|
38 |
def text_to_speech(text):
|
39 |
try:
|
|
|
40 |
max_length = 100 # Set a max length as per model's capability
|
41 |
segments = [text[i:i+max_length] for i in range(0, len(text), max_length)]
|
42 |
audio_paths = []
|
43 |
|
44 |
-
for
|
45 |
inputs = processor(text=segment, return_tensors="pt")
|
46 |
spectrogram = model.generate_speech(inputs["input_ids"], speaker_embeddings)
|
47 |
with torch.no_grad():
|
48 |
speech = vocoder(spectrogram)
|
49 |
-
audio_path = f"speech_segment_{
|
50 |
sf.write(audio_path, speech.numpy(), samplerate=16000)
|
51 |
audio_paths.append(audio_path)
|
52 |
|
@@ -64,28 +65,30 @@ def combine_audio_segments(paths):
|
|
64 |
sf.write("combined_speech.wav", np.array(combined_speech), samplerate)
|
65 |
return "combined_speech.wav"
|
66 |
|
67 |
-
# Text Input
|
68 |
-
text = st.text_area("Type your text
|
69 |
|
|
|
70 |
if st.button("Convert"):
|
71 |
if text:
|
72 |
audio_paths = text_to_speech(text)
|
73 |
combined_audio_path = combine_audio_segments(audio_paths)
|
74 |
-
|
|
|
75 |
st.audio(audio_bytes, format='audio/wav')
|
76 |
else:
|
77 |
st.error("Please enter some text to convert.")
|
78 |
|
79 |
-
# File Uploader
|
80 |
-
uploaded_file = st.file_uploader("Upload
|
81 |
-
|
82 |
if uploaded_file is not None:
|
83 |
stringio = StringIO(uploaded_file.getvalue().decode("utf-8"))
|
84 |
text = stringio.read()
|
85 |
st.write(text)
|
86 |
|
87 |
-
if st.button("Convert Uploaded File", key=
|
88 |
audio_paths = text_to_speech(text)
|
89 |
combined_audio_path = combine_audio_segments(audio_paths)
|
90 |
-
|
91 |
-
|
|
|
|
2 |
import numpy as np
|
3 |
import torch
|
4 |
from transformers import SpeechT5Processor, SpeechT5ForTextToSpeech, SpeechT5HifiGan
|
|
|
5 |
from io import StringIO
|
6 |
+
import soundfile as sf
|
7 |
|
8 |
# Load models outside of function calls for efficiency
|
9 |
@st.cache_data
|
|
|
23 |
|
24 |
speaker_embeddings = get_speaker_embeddings()
|
25 |
|
26 |
+
# Improved Styling
|
27 |
def local_css(file_name):
|
28 |
with open(file_name) as f:
|
29 |
st.markdown(f'<style>{f.read()}</style>', unsafe_allow_html=True)
|
30 |
|
31 |
+
local_css("style.css")
|
32 |
|
33 |
+
# Streamlined Layout
|
34 |
st.title("Text-to-Voice Conversion")
|
35 |
st.markdown("Convert your text to speech using advanced AI models.")
|
36 |
|
37 |
# Function to convert text to speech
|
38 |
def text_to_speech(text):
|
39 |
try:
|
40 |
+
# Segment the text if it's too long
|
41 |
max_length = 100 # Set a max length as per model's capability
|
42 |
segments = [text[i:i+max_length] for i in range(0, len(text), max_length)]
|
43 |
audio_paths = []
|
44 |
|
45 |
+
for segment in segments:
|
46 |
inputs = processor(text=segment, return_tensors="pt")
|
47 |
spectrogram = model.generate_speech(inputs["input_ids"], speaker_embeddings)
|
48 |
with torch.no_grad():
|
49 |
speech = vocoder(spectrogram)
|
50 |
+
audio_path = f"speech_segment_{len(audio_paths)}.wav"
|
51 |
sf.write(audio_path, speech.numpy(), samplerate=16000)
|
52 |
audio_paths.append(audio_path)
|
53 |
|
|
|
65 |
sf.write("combined_speech.wav", np.array(combined_speech), samplerate)
|
66 |
return "combined_speech.wav"
|
67 |
|
68 |
+
# Text Input
|
69 |
+
text = st.text_area("Type your text or upload a text file below.")
|
70 |
|
71 |
+
# Convert Button
|
72 |
if st.button("Convert"):
|
73 |
if text:
|
74 |
audio_paths = text_to_speech(text)
|
75 |
combined_audio_path = combine_audio_segments(audio_paths)
|
76 |
+
audio_file = open(combined_audio_path, 'rb')
|
77 |
+
audio_bytes = audio_file.read()
|
78 |
st.audio(audio_bytes, format='audio/wav')
|
79 |
else:
|
80 |
st.error("Please enter some text to convert.")
|
81 |
|
82 |
+
# File Uploader
|
83 |
+
uploaded_file = st.file_uploader("Upload your text file here", type=['txt'])
|
|
|
84 |
if uploaded_file is not None:
|
85 |
stringio = StringIO(uploaded_file.getvalue().decode("utf-8"))
|
86 |
text = stringio.read()
|
87 |
st.write(text)
|
88 |
|
89 |
+
if st.button("Convert Uploaded File", key=1):
|
90 |
audio_paths = text_to_speech(text)
|
91 |
combined_audio_path = combine_audio_segments(audio_paths)
|
92 |
+
audio_file = open(combined_audio_path, 'rb')
|
93 |
+
audio_bytes = audio_file.read()
|
94 |
+
st.audio(audio_bytes, format='audio/wav')
|