Hugging Face's logo

Spaces:
Translator-team
/
speech-translator
Sleeping

App Files Community
speech-translator
File size: 5,143 Bytes
48b403b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11d2e2e
48b403b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
11d2e2e
48b403b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
import streamlit as st
import whisper
import tempfile
import os
import re
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# =========================
# PAGE CONFIG
# =========================
st.set_page_config(page_title="Speech to Text Translator", page_icon="πŸŽ™οΈ", layout="centered")

# =========================
# UI STYLE
# =========================
st.markdown("""
<style>
html, body, [data-testid="stAppViewContainer"] {
    background: radial-gradient(circle at 20% 20%, #1e293b, #020617 70%);
    color: white;
    font-family: 'Inter', sans-serif;
}
.title {
    text-align: center;
    font-size: 40px;
    font-weight: 700;
}
.subtitle {
    text-align: center;
    color: #cbd5e1;
    margin-bottom: 22px;
}
[data-testid="stFileUploader"] {
    border-radius: 16px;
    border: 1px dashed rgba(255,255,255,0.25);
}
.result-box {
    background: rgba(16,185,129,0.12);
    border-radius: 16px;
    padding: 16px;
    border: 1px solid rgba(16,185,129,0.35);
}
</style>
""", unsafe_allow_html=True)

# =========================
# HEADER
# =========================
st.markdown('<div class="title"> Speech to Text Translator</div>', unsafe_allow_html=True)
st.markdown('<div class="subtitle">Transcribe speech or translate into any language</div>', unsafe_allow_html=True)

# =========================
# TEXT UTILS
# =========================
def split_text(text, max_len=200):
    sentences = re.split(r'(?<=[.!?γ€‚οΌοΌŸ])', text)
    chunks = []
    cur = ""
    for s in sentences:
        if len(cur) + len(s) < max_len:
            cur += " " + s
        else:
            chunks.append(cur.strip())
            cur = s
    if cur:
        chunks.append(cur.strip())
    return chunks

# =========================
# MODEL CACHE
# =========================
@st.cache_resource
def load_models():
    whisper_model = whisper.load_model("base")
    model_name = "facebook/nllb-200-distilled-600M"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    nllb_model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
    return whisper_model, tokenizer, nllb_model

whisper_model, tokenizer, nllb_model = load_models()

# =========================
# LANGUAGE MAP
# =========================
LANG_CODE = {
    "Arabic":"arb_Arab","Assamese":"asm_Beng","Awadhi":"awa_Deva",
    "Bengali":"ben_Beng","Bhojpuri":"bho_Deva","Chinese":"zho_Hans",
    "English":"eng_Latn","French":"fra_Latn","German":"deu_Latn",
    "Hindi":"hin_Deva","Japanese":"jpn_Jpan","Korean":"kor_Hang",
    "Maithili":"mai_Deva","Marathi":"mar_Deva","Persian":"pes_Arab",
    "Punjabi":"pan_Guru","Russian":"rus_Cyrl","Sanskrit":"san_Deva",
    "Spanish":"spa_Latn","Tamil":"tam_Taml","Telugu":"tel_Telu",
    "Urdu":"urd_Arab","Vietnamese":"vie_Latn"
}

# =========================
# OPTIONS
# =========================
col1, col2 = st.columns(2)
with col1:
    transcribe = st.checkbox("πŸ“„ Transcribe")
with col2:
    translate = st.checkbox("🌍 Translate", value=True)

target_lang = None
if translate:
    target_lang = st.selectbox("Translate into", sorted(LANG_CODE.keys()))

# =========================
# UPLOAD
# =========================
audio_file = st.file_uploader(
    "Upload audio (MP3, WAV, M4A, MP4)",
    type=["mp3","wav","m4a","mp4"]
)

if audio_file:
    st.audio(audio_file)

# =========================
# PROCESS
# =========================
if st.button(" Process Audio") and audio_file:

    with st.spinner("Processing audio..."):

        # Save uploaded audio
        with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp:
            tmp.write(audio_file.read())
            tmp_path = tmp.name

        # ---------- TRANSCRIBE ----------
        result = whisper_model.transcribe(tmp_path, fp16=False)
        detected_lang = result["language"]
        text = result["text"]

        st.success(f"Detected language: {detected_lang}")

        output_text = text

        # ---------- TRANSLATE ----------
        if translate and target_lang:
            tgt_code = LANG_CODE[target_lang]
            chunks = split_text(text)

            translated_parts = []

            for chunk in chunks:
                inputs = tokenizer(chunk, return_tensors="pt")

                tokens = nllb_model.generate(
                    **inputs,
                    forced_bos_token_id=tokenizer.convert_tokens_to_ids(tgt_code),
                    max_length=256,
                    num_beams=4,
                    no_repeat_ngram_size=3,
                    repetition_penalty=1.2,
                    early_stopping=True
                )

                translated = tokenizer.batch_decode(tokens, skip_special_tokens=True)[0]
                translated_parts.append(translated)

            output_text = " ".join(translated_parts)

        # ================= OUTPUT =================
        st.markdown("### πŸ“ Output Text")
        st.markdown(f'<div class="result-box">{output_text}</div>', unsafe_allow_html=True)

        st.download_button(
            "⬇ Download Text",
            output_text,
            file_name="translated_text.txt"
        )

        os.remove(tmp_path)