Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import wave
|
3 |
+
import numpy as np
|
4 |
+
from io import BytesIO
|
5 |
+
from huggingface_hub import hf_hub_download
|
6 |
+
from piper import PiperVoice
|
7 |
+
from transformers import pipeline
|
8 |
+
import hazm
|
9 |
+
import typing
|
10 |
+
|
11 |
+
normalizer = hazm.Normalizer()
|
12 |
+
sent_tokenizer = hazm.SentenceTokenizer()
|
13 |
+
word_tokenizer = hazm.WordTokenizer()
|
14 |
+
|
15 |
+
tagger_path = hf_hub_download(repo_id="gyroing/HAZM_POS_TAGGER", filename="pos_tagger.model")
|
16 |
+
tagger = hazm.POSTagger(model=tagger_path)
|
17 |
+
model_path = hf_hub_download(repo_id="gyroing/Persian-Piper-Model-gyro", filename="fa_IR-gyro-medium.onnx")
|
18 |
+
config_path = hf_hub_download(repo_id="gyroing/Persian-Piper-Model-gyro", filename="fa_IR-gyro-medium.onnx.json")
|
19 |
+
voice = PiperVoice.load(model_path, config_path)
|
20 |
+
|
21 |
+
def preprocess_text(text: str) -> typing.List[typing.List[str]]:
|
22 |
+
"""Split/normalize text into sentences/words with hazm"""
|
23 |
+
text = normalizer.normalize(text)
|
24 |
+
processed_sentences = []
|
25 |
+
|
26 |
+
for sentence in sent_tokenizer.tokenize(text):
|
27 |
+
words = word_tokenizer.tokenize(sentence)
|
28 |
+
processed_words = fix_words(words)
|
29 |
+
processed_sentences.append(" ".join(processed_words))
|
30 |
+
return " ".join(processed_sentences)
|
31 |
+
def fix_words(words: typing.List[str]) -> typing.List[str]:
|
32 |
+
fixed_words = []
|
33 |
+
|
34 |
+
for word, pos in tagger.tag(words):
|
35 |
+
if pos[-1] == "Z":
|
36 |
+
if word[-1] != "ِ":
|
37 |
+
if (word[-1] == "ه") and (word[-2] != "ا"):
|
38 |
+
word += "ی"
|
39 |
+
word += "ِ"
|
40 |
+
|
41 |
+
|
42 |
+
fixed_words.append(word)
|
43 |
+
|
44 |
+
return fixed_words
|
45 |
+
|
46 |
+
def synthesize_speech(text):
|
47 |
+
|
48 |
+
|
49 |
+
# Create an in-memory buffer for the WAV file
|
50 |
+
buffer = BytesIO()
|
51 |
+
with wave.open(buffer, 'wb') as wav_file:
|
52 |
+
wav_file.setframerate(voice.config.sample_rate)
|
53 |
+
wav_file.setsampwidth(2) # 16-bit
|
54 |
+
wav_file.setnchannels(1) # mono
|
55 |
+
|
56 |
+
# Synthesize speech
|
57 |
+
eztext = preprocess_text(text)
|
58 |
+
voice.synthesize(eztext, wav_file)
|
59 |
+
|
60 |
+
# Convert buffer to NumPy array for Gradio output
|
61 |
+
buffer.seek(0)
|
62 |
+
audio_data = np.frombuffer(buffer.read(), dtype=np.int16)
|
63 |
+
|
64 |
+
return audio_data.tobytes()
|
65 |
+
|
66 |
+
# Using Gradio Blocks
|
67 |
+
with gr.Blocks(theme=gr.themes.Base()) as blocks:
|
68 |
+
input_text = gr.Textbox(label="Input")
|
69 |
+
output_audio = gr.Audio(label="Output", type="numpy")
|
70 |
+
submit_button = gr.Button("Synthesize")
|
71 |
+
|
72 |
+
submit_button.click(synthesize_speech, inputs=input_text, outputs=[output_audio])
|
73 |
+
# Run the app
|
74 |
+
blocks.launch()
|