Kuznetsov AV
commited on
Commit
•
6feeeab
1
Parent(s):
c910ab2
text-to-speech module completed
Browse files- kuznetsov_av/__init__.py +0 -0
- kuznetsov_av/kuznetsov_av.py +0 -23
- kuznetsov_av/requirements.txt +0 -4
- kuznetsov_av/text_to_speech_converter.py +41 -0
- requirements.txt +3 -2
- run.py +9 -1
kuznetsov_av/__init__.py
ADDED
File without changes
|
kuznetsov_av/kuznetsov_av.py
DELETED
@@ -1,23 +0,0 @@
|
|
1 |
-
from transformers import pipeline
|
2 |
-
from datasets import load_dataset
|
3 |
-
import torch
|
4 |
-
import streamlit as st
|
5 |
-
|
6 |
-
@st.cache_resource
|
7 |
-
def load_model():
|
8 |
-
synthesiser = pipeline("text-to-speech", "microsoft/speecht5_tts")
|
9 |
-
|
10 |
-
embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
|
11 |
-
speaker_embedding = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
|
12 |
-
|
13 |
-
return synthesiser, speaker_embedding
|
14 |
-
|
15 |
-
synthesiser, speaker_embedding = load_model()
|
16 |
-
|
17 |
-
text = st.text_area('Enter English text here')
|
18 |
-
st.write(f'You wrote {len(text)} characters.')
|
19 |
-
|
20 |
-
if st.button('Speech'):
|
21 |
-
speech = synthesiser(text, forward_params={"speaker_embeddings": speaker_embedding})
|
22 |
-
|
23 |
-
st.audio(speech['audio'], sample_rate=speech['sampling_rate'])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
kuznetsov_av/requirements.txt
DELETED
@@ -1,4 +0,0 @@
|
|
1 |
-
datasets==2.14.6
|
2 |
-
streamlit==1.28.1
|
3 |
-
torch==2.1.0
|
4 |
-
transformers==4.35.0
|
|
|
|
|
|
|
|
|
|
kuznetsov_av/text_to_speech_converter.py
ADDED
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import pipeline
|
2 |
+
import transformers.pipelines.text_to_audio
|
3 |
+
from datasets import load_dataset
|
4 |
+
import datasets.arrow_dataset
|
5 |
+
import torch
|
6 |
+
import numpy as np
|
7 |
+
|
8 |
+
|
9 |
+
def load_model() -> transformers.pipelines.text_to_audio.TextToAudioPipeline:
|
10 |
+
"""
|
11 |
+
Подгрузка модели преобразования текста в речь
|
12 |
+
:return: class TextToAudioPipeline
|
13 |
+
"""
|
14 |
+
return pipeline("text-to-speech", "microsoft/speecht5_tts")
|
15 |
+
|
16 |
+
|
17 |
+
def load_speaker_dataset() -> datasets.arrow_dataset.Dataset:
|
18 |
+
"""
|
19 |
+
Подгрузка датасета для озвучивания текста
|
20 |
+
:return: class Dataset
|
21 |
+
"""
|
22 |
+
return load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")
|
23 |
+
|
24 |
+
|
25 |
+
def text_to_speech(
|
26 |
+
text: str,
|
27 |
+
synthesiser: transformers.pipelines.text_to_audio.TextToAudioPipeline,
|
28 |
+
embeddings_dataset: datasets.arrow_dataset.Dataset
|
29 |
+
) -> (np.ndarray, int):
|
30 |
+
"""
|
31 |
+
Преобразование текста в речь
|
32 |
+
:param text: Текст
|
33 |
+
:param synthesiser: pipeline для озвучивания текста
|
34 |
+
:param embeddings_dataset: dataset для озвучивания текста
|
35 |
+
:return: tuple (audio data, sampling rate)
|
36 |
+
"""
|
37 |
+
speaker_embedding = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)
|
38 |
+
|
39 |
+
speech = synthesiser(text, forward_params={"speaker_embeddings": speaker_embedding})
|
40 |
+
|
41 |
+
return speech['audio'], speech['sampling_rate']
|
requirements.txt
CHANGED
@@ -1,6 +1,7 @@
|
|
1 |
datasets==2.14.6
|
|
|
2 |
streamlit==1.28.1
|
3 |
torch==2.1.0
|
4 |
transformers==4.35.0
|
5 |
-
sentencepiece
|
6 |
-
sacremoses
|
|
|
1 |
datasets==2.14.6
|
2 |
+
numpy==1.26.2
|
3 |
streamlit==1.28.1
|
4 |
torch==2.1.0
|
5 |
transformers==4.35.0
|
6 |
+
sentencepiece==0.1.99
|
7 |
+
sacremoses==0.1.1
|
run.py
CHANGED
@@ -2,9 +2,12 @@ import streamlit as st
|
|
2 |
|
3 |
from mulyavin_aa import langdetector
|
4 |
from mulyavin_aa import translator
|
|
|
5 |
|
6 |
LANG_DETECTOR = "LANG_DETECTOR"
|
7 |
TRANSLATOR = "TRANSLATOR"
|
|
|
|
|
8 |
|
9 |
|
10 |
@st.cache_resource
|
@@ -16,6 +19,8 @@ def load_models() -> dict:
|
|
16 |
models = dict()
|
17 |
models[LANG_DETECTOR] = langdetector.load_text_detection_model()
|
18 |
models[TRANSLATOR] = translator.load_text_translator_model()
|
|
|
|
|
19 |
|
20 |
return models
|
21 |
|
@@ -49,7 +54,10 @@ def main_app():
|
|
49 |
tab1, tab2, tab3 = st.tabs(['Озвученный текст', 'Таб 2', 'Таб 3'])
|
50 |
with tab1:
|
51 |
st.header("Озвученный текст на английском языке")
|
52 |
-
#
|
|
|
|
|
|
|
53 |
|
54 |
with tab2:
|
55 |
st.header("Таб 2")
|
|
|
2 |
|
3 |
from mulyavin_aa import langdetector
|
4 |
from mulyavin_aa import translator
|
5 |
+
from kuznetsov_av import text_to_speech_converter
|
6 |
|
7 |
LANG_DETECTOR = "LANG_DETECTOR"
|
8 |
TRANSLATOR = "TRANSLATOR"
|
9 |
+
TEXT_TO_SPEECH = "TEXT_TO_SPEECH"
|
10 |
+
SPEAKER_DATASET = "SPEAKER_DATASET"
|
11 |
|
12 |
|
13 |
@st.cache_resource
|
|
|
19 |
models = dict()
|
20 |
models[LANG_DETECTOR] = langdetector.load_text_detection_model()
|
21 |
models[TRANSLATOR] = translator.load_text_translator_model()
|
22 |
+
models[TEXT_TO_SPEECH] = text_to_speech_converter.load_model()
|
23 |
+
models[SPEAKER_DATASET] = text_to_speech_converter.load_speaker_dataset()
|
24 |
|
25 |
return models
|
26 |
|
|
|
54 |
tab1, tab2, tab3 = st.tabs(['Озвученный текст', 'Таб 2', 'Таб 3'])
|
55 |
with tab1:
|
56 |
st.header("Озвученный текст на английском языке")
|
57 |
+
# Преобразование текста в речь
|
58 |
+
audio_data, sampling_rate = text_to_speech_converter.text_to_speech(
|
59 |
+
input_text, models[TEXT_TO_SPEECH], models[SPEAKER_DATASET])
|
60 |
+
st.audio(data=audio_data, sample_rate=sampling_rate)
|
61 |
|
62 |
with tab2:
|
63 |
st.header("Таб 2")
|