app.v2.py

import gradio as gr
import librosa
import soundfile
import tempfile
import os
import uuid
import json
import re

from nemo.collections.asr.models import ASRModel
from nemo.utils import logging

from align import main, AlignmentConfig, ASSFileConfig


SAMPLE_RATE = 16000

logging.setLevel(logging.INFO)

# --- ASS TO SRT/LRC/ELRC CONVERTER FUNCTIONS ---

def format_srt_time(secs):
    h = int(secs // 3600)
    m = int((secs % 3600) // 60)
    s = int(secs % 60)
    ms = int(round((secs % 1) * 1000))
    if ms == 1000:
        ms = 0
        s += 1
    return f"{h:02d}:{m:02d}:{s:02d},{ms:03d}"

def format_lrc_time(secs):
    m = int(secs // 60)
    s = int(secs % 60)
    cs = int(round((secs % 1) * 100))
    if cs == 100:
        cs = 0
        s += 1
    return f"{m:02d}:{s:02d}.{cs:02d}"

def build_color_map(raw_text):
    color_map = []
    current_color = "default"
    regex = re.compile(r'(\{[^}]+\})|([^{]+)')
    for match in regex.finditer(raw_text):
        tag = match.group(1)
        text = match.group(2)
        if tag:
            c_match = re.search(r'\\c&H([0-9a-fA-F]+)&', tag, re.IGNORECASE)
            if c_match:
                current_color = c_match.group(1).lower()
        elif text:
            for _ in text:
                color_map.append(current_color)
    return color_map

def parse_ass_to_segments(text):
    lines = text.split('\n')
    dialogues = []
    
    for line in lines:
        if line.strip().startswith('Dialogue:'):
            parts = line[10:].split(',')
            start = parse_ass_time(parts[1])
            end = parse_ass_time(parts[2])
            raw_text = ','.join(parts[9:])
            clean_text_raw = re.sub(r'\{[^}]+\}', '', raw_text)
            clean_text = re.sub(r'\\N', ' ', clean_text_raw)
            clean_text = re.sub(r'\s+', ' ', clean_text).strip()
            dialogues.append({
                'start': start, 'end': end, 
                'rawText': raw_text, 
                'cleanTextRaw': clean_text_raw, 
                'cleanText': clean_text
            })

    segments = []
    current_segment = None

    for d in dialogues:
        if not current_segment or current_segment['cleanText'] != d['cleanText']:
            if current_segment:
                segments.append(current_segment)
            current_segment = {
                'cleanText': d['cleanText'],
                'cleanTextRaw': d['cleanTextRaw'],
                'startTime': d['start'],
                'endTime': d['end'],
                'slices': [d]
            }
        else:
            current_segment['slices'].append(d)
            current_segment['endTime'] = d['end']
            
    if current_segment:
        segments.append(current_segment)

    for seg in segments:
        if not seg['slices']:
            continue
        base_color_map = build_color_map(seg['slices'][0]['rawText'])
        if not base_color_map:
            seg['words'] = []
            continue
            
        base_color = base_color_map[-1] 

        words = []
        for match in re.finditer(r'\S+', seg['cleanTextRaw']):
            txt = match.group(0).replace('\\N', '')
            if txt.strip():
                words.append({
                    'text': txt,
                    'startIndex': match.start(),
                    'isFirstOfSegment': len(words) == 0
                })

        for w in words:
            char_pos = w['startIndex']
            c0 = base_color_map[char_pos] if char_pos < len(base_color_map) else "default"
            w_start = seg['startTime']
            w_end = seg['endTime']

            transitions = []
            prev_color = c0

            for i in range(1, len(seg['slices'])):
                slice_d = seg['slices'][i]
                cmap = build_color_map(slice_d['rawText'])
                c_curr = cmap[char_pos] if char_pos < len(cmap) else "default"
                if c_curr != prev_color:
                    transitions.append({'time': slice_d['start'], 'color': c_curr})
                    prev_color = c_curr

            if c0 == base_color:
                w_start = transitions[0]['time'] if len(transitions) > 0 else seg['startTime']
                w_end = transitions[1]['time'] if len(transitions) > 1 else seg['endTime']
            else:
                w_start = seg['startTime']
                w_end = transitions[0]['time'] if len(transitions) > 0 else seg['endTime']

            w['startTime'] = w_start
            w['endTime'] = w_end
            
        # --- CRITICAL FIX ---
        # Force the first word and last word to definitively stretch to the overarching segment boundaries
        if words:
            words[0]['startTime'] = seg['startTime']
            words[-1]['endTime'] = seg['endTime']

        seg['words'] = words

    return segments

def generate_srt_segments(segments):
    out = []
    counter = 1
    for seg in segments:
        out.append(str(counter))
        out.append(f"{format_srt_time(seg['startTime'])} --> {format_srt_time(seg['endTime'])}")
        out.append(seg['cleanText'])
        out.append("")
        counter += 1
    return "\n".join(out).strip()

def generate_srt_word_by_word(segments):
    out = []
    counter = 1
    for seg in segments:
        for w in seg['words']:
            out.append(str(counter))
            out.append(f"{format_srt_time(w['startTime'])} --> {format_srt_time(w['endTime'])}")
            out.append(w['text'])
            out.append("")
            counter += 1
    return "\n".join(out).strip()

def generate_srt_additive(segments):
    out = []
    counter = 1
    for seg in segments:
        accumulated = []
        for i, w in enumerate(seg['words']):
            out.append(str(counter))
            actual_start = w['startTime'] if i == 0 else seg['words'][i-1]['endTime']
            out.append(f"{format_srt_time(actual_start)} --> {format_srt_time(w['endTime'])}")
            accumulated.append(w['text'])
            out.append(" ".join(accumulated))
            out.append("")
            counter += 1
    return "\n".join(out).strip()

def generate_lrc(segments):
    out = []
    for seg in segments:
        out.append(f"[{format_lrc_time(seg['startTime'])}]{seg['cleanText']}")
        out.append(f"[{format_lrc_time(seg['endTime'])}]")
    return "\n".join(out).strip()

def generate_elrc(segments):
    out = []
    for seg in segments:
        line = f"[{format_lrc_time(seg['startTime'])}]"
        for i, w in enumerate(seg['words']):
            if i == 0:
                line += f"<{format_lrc_time(w['startTime'])}>{w['text']}"
            else:
                line += f" <{format_lrc_time(w['startTime'])}>{w['text']}"
        out.append(line)
        out.append(f"[{format_lrc_time(seg['endTime'])}]")
    return "\n".join(out).strip()

# --- ORIGINAL PARSING FUNCTIONS ---

def parse_srt(content):
    pattern = re.compile(r'\d+\n(\d{2}:\d{2}:\d{2},\d{3}) --> (\d{2}:\d{2}:\d{2},\d{3})\n((?:(?!\n\n).)*)', re.DOTALL)
    matches = pattern.findall(content + "\n\n") 
    segments = []
    
    def time_to_sec(t_str):
        h, m, s_ms = t_str.split(':')
        s, ms = s_ms.split(',')
        return int(h) * 3600 + int(m) * 60 + int(s) + int(ms) / 1000.0
        
    for match in matches:
        start = time_to_sec(match[0])
        end = time_to_sec(match[1])
        text = match[2].replace('\n', ' ').strip()
        segments.append({"start": start, "end": end, "text": text})
        
    return segments

def parse_lrc(content, audio_duration):
    lines = content.split('\n')
    pattern = re.compile(r'\[(\d{2}):(\d{2}\.\d{2,3})\](.*)')
    raw_markers = []
    
    for line in lines:
        match = pattern.match(line.strip())
        if match:
            m, s, text = match.groups()
            start = int(m) * 60 + float(s)
            raw_markers.append({"start": start, "text": text.strip()})
            
    segments = []
    for i in range(len(raw_markers)):
        current = raw_markers[i]
        text = current["text"]
        
        if text and text != "#":
            end_time = audio_duration
            if i + 1 < len(raw_markers):
                end_time = raw_markers[i+1]["start"]
                
            if end_time > current["start"]:
                segments.append({
                    "start": current["start"], 
                    "end": end_time, 
                    "text": text
                })
                
    return segments

def parse_ass_time(t_str):
    h, m, s = t_str.strip().split(':')
    return float(h) * 3600 + float(m) * 60 + float(s)

def format_ass_time(sec):
    sec = max(0.0, sec)
    nh = int(sec // 3600)
    nm = int((sec % 3600) // 60)
    ns = sec % 60
    return f"{nh:d}:{nm:02d}:{ns:05.2f}"

# -----------------------------

def get_audio_data_and_duration(file):
    data, sr = librosa.load(file)

    if sr != SAMPLE_RATE:
        data = librosa.resample(data, orig_sr=sr, target_sr=SAMPLE_RATE)

    # monochannel
    data = librosa.to_mono(data)

    duration = librosa.get_duration(y=data, sr=SAMPLE_RATE)
    return data, duration


def get_char_tokens(text, model):
    tokens = []
    for character in text:
        if character in model.decoder.vocabulary:
            tokens.append(model.decoder.vocabulary.index(character))
        else:
            tokens.append(len(model.decoder.vocabulary))  # return unk token (same as blank token)

    return tokens


def get_S_prime_and_T(text, model_name, model, audio_duration):
    if "citrinet" in model_name or "_fastconformer_" in model_name:
        output_timestep_duration = 0.08
    elif "_conformer_" in model_name:
        output_timestep_duration = 0.04
    elif "quartznet" in model_name:
        output_timestep_duration = 0.02
    else:
        raise RuntimeError("unexpected model name")

    T = int(audio_duration / output_timestep_duration) + 1

    if hasattr(model, 'tokenizer'):
        all_tokens = model.tokenizer.text_to_ids(text)
    elif hasattr(model.decoder, "vocabulary"): 
        all_tokens = get_char_tokens(text, model)
    else:
        raise RuntimeError("cannot obtain tokens from this model")

    n_token_repetitions = 0
    for i_tok in range(1, len(all_tokens)):
        if all_tokens[i_tok] == all_tokens[i_tok - 1]:
            n_token_repetitions += 1

    S_prime = len(all_tokens) + n_token_repetitions

    return S_prime, T


def delete_mp4s_except_given_filepath(filepath):
    files_in_dir = os.listdir()
    mp4_files_in_dir = [x for x in files_in_dir if x.endswith(".mp4")]
    for mp4_file in mp4_files_in_dir:
        if mp4_file != filepath:
            os.remove(mp4_file)


def align(Microphone, File_Upload, subs_file, text, split_on_newline, progress=gr.Progress()):
    utt_id = uuid.uuid4()
    output_video_filepath = f"{utt_id}.mp4"
    delete_mp4s_except_given_filepath(output_video_filepath)

    output_info = ""
    ass_text = ""

    progress(0, desc="Validating input")

    if (Microphone is not None) and (File_Upload is not None):
        raise gr.Error("Please use either the microphone or file upload input - not both")
    elif (Microphone is None) and (File_Upload is None):
        raise gr.Error("You have to either use the microphone or upload an audio file")
    elif Microphone is not None:
        file = Microphone
    else:
        file = File_Upload

    audio_data, duration = get_audio_data_and_duration(file)

    progress(0.1, desc="Loading speech recognition model")
    model_name = "ayymen/stt_zgh_fastconformer_ctc_small"
    model = ASRModel.from_pretrained(model_name)

    segments = []
    if subs_file is not None:
        with open(subs_file.name, 'r', encoding='utf-8') as f:
            subs_content = f.read()
        
        if subs_file.name.lower().endswith('.srt'):
            segments = parse_srt(subs_content)
        elif subs_file.name.lower().endswith('.lrc'):
            segments = parse_lrc(subs_content, duration)
        else:
            raise gr.Error("Subtitle file must be an .srt or .lrc file.")
            
    with tempfile.TemporaryDirectory() as tmpdir:
        manifest_path = os.path.join(tmpdir, f"{utt_id}_manifest.json")
        
        if segments:
            progress(0.2, desc="Chunking audio and generating manifest")
            with open(manifest_path, 'w', encoding='utf-8') as fout:
                for i, seg in enumerate(segments):
                    S_prime, T = get_S_prime_and_T(seg['text'], model_name, model, seg['end'] - seg['start'])
                    if S_prime > T:
                        raise gr.Error(f"Segment {i} ('{seg['text']}') has too much text for its audio duration ({seg['end'] - seg['start']:.2f}s).")

                    start_sample = int(seg['start'] * SAMPLE_RATE)
                    end_sample = int(seg['end'] * SAMPLE_RATE)
                    chunk_data = audio_data[start_sample:end_sample]

                    chunk_path = os.path.join(tmpdir, f"{utt_id}_{i:04d}.wav")
                    soundfile.write(chunk_path, chunk_data, SAMPLE_RATE)

                    seg_text = seg['text'].replace('\n', '|') if split_on_newline else seg['text'].replace('\n', ' ')

                    data = {
                        "audio_filepath": chunk_path,
                        "text": seg_text,
                    }
                    fout.write(f"{json.dumps(data)}\n")
                    
            resegment_text_to_fill_space = False

        else:
            audio_path = os.path.join(tmpdir, f'{utt_id}.wav')
            soundfile.write(audio_path, audio_data, SAMPLE_RATE)

            if not text:
                progress(0.2, desc="Transcribing audio")
                text = model.transcribe([audio_path])[0]
                if 'hybrid' in model_name:
                    text = text[0]

                if text == "":
                    raise gr.Error("ERROR: the ASR model did not detect any speech. Please upload audio with speech.")

                output_info += (
                    "You did not enter any input text, so the ASR model's transcription will be used:\n"
                    "--------------------------\n"
                    f"{text}\n"
                    "--------------------------\n"
                    f"You could try pasting the transcription into the text input box, correcting any"
                    " transcription errors, and clicking 'Submit' again."
                )

            if split_on_newline:
                text = "|".join(list(filter(None, text.split("\n"))))
                
            S_prime, T = get_S_prime_and_T(text, model_name, model, duration)
            if S_prime > T:
                raise gr.Error("The number of tokens in the input text is too long compared to the duration of the audio.")

            with open(manifest_path, 'w', encoding='utf-8') as fout:
                data = {
                    "audio_filepath": audio_path,
                    "text": text,
                }
                fout.write(f"{json.dumps(data)}\n")

            resegment_text_to_fill_space = "|" not in text


        alignment_config = AlignmentConfig(
            pretrained_name=model_name,
            manifest_filepath=manifest_path,
            output_dir=f"{tmpdir}/nfa_output/",
            audio_filepath_parts_in_utt_id=1,
            batch_size=1,
            use_local_attention=True,
            additional_segment_grouping_separator="|",
            save_output_file_formats=["ass", "ctm"],
            ass_file_config=ASSFileConfig(
                fontsize=45,
                resegment_text_to_fill_space=resegment_text_to_fill_space,
                max_lines_per_segment=4,
            ),
        )

        progress(0.5, desc="Aligning audio")
        main(alignment_config)
        progress(0.95, desc="Saving generated alignments")


        ass_path = "word_level.ass"
        word_ctm_path = "word_level.ctm"
        segment_ctm_path = "segment_level.ctm"
        
        if segments:
            # --- ASS MERGING ---
            merged_ass = ""
            header_written = False
            
            for i, seg in enumerate(segments):
                chunk_ass_path = f"{tmpdir}/nfa_output/ass/words/{utt_id}_{i:04d}.ass"
                if os.path.exists(chunk_ass_path):
                    chunk_lines = []
                    
                    with open(chunk_ass_path, "r", encoding='utf-8') as f:
                        for line in f:
                            if line.startswith("Dialogue:"):
                                parts = line.split(",", 9)
                                if len(parts) >= 10:
                                    chunk_lines.append(parts)
                            elif not header_written:
                                merged_ass += line
                    header_written = True
                    
                    if chunk_lines:
                        for j, parts in enumerate(chunk_lines):
                            local_start = parse_ass_time(parts[1])
                            local_end = parse_ass_time(parts[2])
                            
                            global_start = local_start + seg['start']
                            global_end = local_end + seg['start']
                            
                            if j == 0:
                                global_start = seg['start']
                                
                            if j == len(chunk_lines) - 1:
                                global_end = seg['end']
                                
                                if i < len(segments) - 1:
                                    next_start = segments[i+1]['start']
                                    if global_end >= next_start:
                                        global_end = next_start - 0.05
                            
                            if global_start >= global_end:
                                global_start = global_end - 0.01

                            parts[1] = format_ass_time(global_start)
                            parts[2] = format_ass_time(global_end)
                            
                            merged_ass += ",".join(parts)
            
            with open(ass_path, "w", encoding="utf-8") as f:
                f.write(merged_ass)
                
            ass_text = merged_ass

            # --- CTM MERGING (Now strictly bounded to segment edges too) ---
            for ctm_type, out_path in [("words", word_ctm_path), ("segments", segment_ctm_path)]:
                merged_ctm = ""
                for i, seg in enumerate(segments):
                    chunk_ctm_path = f"{tmpdir}/nfa_output/ctm/{ctm_type}/{utt_id}_{i:04d}.ctm"
                    if os.path.exists(chunk_ctm_path):
                        lines = []
                        with open(chunk_ctm_path, "r", encoding='utf-8') as f:
                            for line in f:
                                parts = line.strip().split()
                                if len(parts) >= 5:
                                    lines.append(parts)
                        
                        for j, parts in enumerate(lines):
                            parts[0] = str(utt_id)
                            l_start = float(parts[2])
                            l_dur = float(parts[3])
                            l_end = l_start + l_dur
                            
                            g_start = l_start + seg['start']
                            g_end = l_end + seg['start']
                            
                            if ctm_type == "segments":
                                g_start = seg['start']
                                g_end = seg['end']
                                if i < len(segments) - 1:
                                    next_start = segments[i+1]['start']
                                    if g_end >= next_start:
                                        g_end = next_start - 0.05
                            else: # words
                                if j == 0:
                                    g_start = seg['start']
                                if j == len(lines) - 1:
                                    g_end = seg['end']
                                    if i < len(segments) - 1:
                                        next_start = segments[i+1]['start']
                                        if g_end >= next_start:
                                            g_end = next_start - 0.05
                                            
                            if g_start >= g_end:
                                g_start = g_end - 0.01
                                
                            parts[2] = f"{g_start:.2f}"
                            parts[3] = f"{(g_end - g_start):.2f}"
                            merged_ctm += " ".join(parts) + "\n"
                
                with open(out_path, "w", encoding="utf-8") as f:
                    f.write(merged_ctm)

        else:
            ass_file_for_video = f"{tmpdir}/nfa_output/ass/words/{utt_id}.ass"
            with open(ass_file_for_video, "r", encoding="utf-8") as f:
                ass_text = f.read()
            with open(ass_path, "w", encoding="utf-8") as f:
                f.write(ass_text)

            with open(f"{tmpdir}/nfa_output/ctm/words/{utt_id}.ctm", "r", encoding="utf-8") as f:
                with open(word_ctm_path, "w", encoding="utf-8") as out_f:
                    out_f.write(f.read())

            with open(f"{tmpdir}/nfa_output/ctm/segments/{utt_id}.ctm", "r", encoding="utf-8") as f:
                with open(segment_ctm_path, "w", encoding="utf-8") as out_f:
                    out_f.write(f.read())
                    

        # --- GENERATE ADDITIONAL SUBTITLE FORMATS ---
        segments_for_subs = parse_ass_to_segments(ass_text)

        srt_seg_path = "segments.srt"
        with open(srt_seg_path, "w", encoding="utf-8") as f:
            f.write(generate_srt_segments(segments_for_subs))

        srt_word_path = "word_by_word.srt"
        with open(srt_word_path, "w", encoding="utf-8") as f:
            f.write(generate_srt_word_by_word(segments_for_subs))

        srt_add_path = "additive.srt"
        with open(srt_add_path, "w", encoding="utf-8") as f:
            f.write(generate_srt_additive(segments_for_subs))

        lrc_path = "segments.lrc"
        with open(lrc_path, "w", encoding="utf-8") as f:
            f.write(generate_lrc(segments_for_subs))

        elrc_path = "word_level.elrc"
        with open(elrc_path, "w", encoding="utf-8") as f:
            f.write(generate_elrc(segments_for_subs))

        # --- VIDEO GENERATION ---
        full_audio_path = os.path.join(tmpdir, "full_audio.wav")
        soundfile.write(full_audio_path, audio_data, SAMPLE_RATE)

        ffmpeg_command = (
            f"ffmpeg -y -i {full_audio_path} "
            "-f lavfi -i color=c=white:s=1280x720:r=50 "
            "-crf 1 -shortest -vcodec libx264 -pix_fmt yuv420p "
            f"-vf \"ass='{ass_path}'\" "
            f"{output_video_filepath}"
        )
        os.system(ffmpeg_command)

    return (
        output_video_filepath, 
        gr.update(value=output_info, visible=True if output_info else False), 
        output_video_filepath, 
        gr.update(value=ass_path, visible=True), 
        gr.update(value=word_ctm_path, visible=True), 
        gr.update(value=segment_ctm_path, visible=True),
        gr.update(value=srt_seg_path, visible=True),
        gr.update(value=srt_word_path, visible=True),
        gr.update(value=srt_add_path, visible=True),
        gr.update(value=lrc_path, visible=True),
        gr.update(value=elrc_path, visible=True)
    )


def delete_non_tmp_video(video_path):
    if video_path:
        if os.path.exists(video_path):
            os.remove(video_path)
    return None


with gr.Blocks(title="NeMo Forced Aligner", theme="huggingface") as demo:
    non_tmp_output_video_filepath = gr.State([])

    with gr.Row():
        with gr.Column():
            gr.Markdown("# NeMo Forced Aligner")
            gr.Markdown(
                "Demo for [NeMo Forced Aligner](https://github.com/NVIDIA/NeMo/tree/main/tools/nemo_forced_aligner) (NFA). "
                "Upload audio in Tamazight and (optionally) the text spoken in the audio to generate a video where each part of the text will be highlighted as it is spoken. "
                "**Now supports syncing with pre-timed SRT or LRC files and generates all subtitle formats instantly!**",
            )

    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown("## Input")
            mic_in = gr.Audio(sources=["microphone"], type='filepath', label="Microphone input")
            audio_file_in = gr.Audio(sources=["upload"], type='filepath', label="File upload")
            
            subs_file_in = gr.File(label="[Optional] Upload an SRT or LRC file to constrain alignment to predefined timestamps", file_types=[".srt", ".lrc"])
            
            ref_text = gr.Textbox(
                label="[Optional] The reference text. Use '|' separators to specify which text will appear together. "
                "Leave this field blank to use an ASR model's transcription as the reference text instead. (Ignored if SRT/LRC is uploaded)"
            )
            split_on_newline = gr.Checkbox(
                True,
                label="Separate text on new lines",
            )

            submit_button = gr.Button("Submit")

        with gr.Column(scale=1):
            gr.Markdown("## Output Video")
            video_out = gr.Video(label="Output Video")
            text_out = gr.Textbox(label="Output Info", visible=False)
            
            gr.Markdown("## Download Subtitle Files")
            with gr.Row():
                ass_file = gr.File(label="ASS (Karaoke)", visible=False)
                word_ctm_file = gr.File(label="CTM (Word-level)", visible=False)
                segment_ctm_file = gr.File(label="CTM (Segment-level)", visible=False)
                
            with gr.Row():
                srt_seg_file = gr.File(label="SRT (Segments)", visible=False)
                srt_word_file = gr.File(label="SRT (Word-by-Word)", visible=False)
                srt_add_file = gr.File(label="SRT (Additive)", visible=False)
                
            with gr.Row():
                lrc_file = gr.File(label="LRC (Segments)", visible=False)
                elrc_file = gr.File(label="ELRC (Word-level)", visible=False)

    with gr.Row():
        gr.HTML(
            "<p style='text-align: center'>"
                "Tutorial: <a href='https://colab.research.google.com/github/NVIDIA/NeMo/blob/main/tutorials/tools/NeMo_Forced_Aligner_Tutorial.ipynb' target='_blank'>\"How to use NFA?\"</a> 🚀 | "
                "Blog post: <a href='https://nvidia.github.io/NeMo/blogs/2023/2023-08-forced-alignment/' target='_blank'>\"How does forced alignment work?\"</a> 📚 | "
                "NFA <a href='https://github.com/NVIDIA/NeMo/tree/main/tools/nemo_forced_aligner/' target='_blank'>Github page</a> 👩‍💻"
            "</p>"
        )

    submit_button.click(
        fn=align,
        inputs=[mic_in, audio_file_in, subs_file_in, ref_text, split_on_newline],
        outputs=[
            video_out, text_out, non_tmp_output_video_filepath, 
            ass_file, word_ctm_file, segment_ctm_file,
            srt_seg_file, srt_word_file, srt_add_file, lrc_file, elrc_file
        ],
    ).then(
        fn=delete_non_tmp_video, inputs=[non_tmp_output_video_filepath], outputs=None,
    )
    
    example_2 = """ⵜⴰⴽⵟⵟⵓⵎⵜ ⵏ ⵜⵙⴰⴷⵓⴼⵜ.
ⵙ ⵉⵙⵎ ⵏ ⵕⴱⴱⵉ ⴰⵎⴰⵍⵍⴰⵢ ⴰⵎⵙⵎⵓⵍⵍⵓ.
ⴰⵎⵓⵢ ⵉ ⵕⴱⴱⵉ ⵍⵍⵉ ⵎⵓ ⵜⴳⴰ ⵜⵓⵍⵖⵉⵜ ⵜⵉⵏⵏⵙ, ⵕⴱⴱⵉ ⵏ ⵉⵖⵥⵡⴰⵕⵏ, ⴽⵔⴰ ⴳⴰⵏ.
ⴰⵎⴰⵍⵍⴰⵢ ⴰⵎⵙⵎⵓⵍⵍⵓ, ⵖ ⵜⵎⵣⵡⴰⵔⵓⵜ ⵓⵍⴰ ⵖ ⵜⵎⴳⴳⴰⵔⵓⵜ.
ⴰⴳⵍⵍⵉⴷ ⵏ ⵡⴰⵙⵙ ⵏ ⵓⴼⵔⴰ, ⴰⵙⵙ ⵏ ⵓⵙⵙⵃⵙⵓ, ⴽⵔⴰⵉⴳⴰⵜ ⵢⴰⵏ ⴷ ⵎⴰⴷ ⵉⵙⴽⵔ.
ⵀⴰ ⵏⵏ ⴽⵢⵢⵉ ⴽⴰ ⵙ ⵏⵙⵙⵓⵎⴷ, ⴷ ⴽⵢⵢⵉ ⴽⴰ ⴰⴷ ⵏⵎⵎⵜⵔ.
ⵙⵎⵓⵏ ⴰⵖ, ⵜⵎⵍⵜ ⴰⵖ, ⴰⵖⴰⵔⴰⵙ ⵢⵓⵖⴷⵏ.
ⴰⵖⴰⵔⴰⵙ ⵏ ⵖⵡⵉⵍⵍⵉ ⵜⵙⵏⵏⵓⴼⴰⵜ, ⵓⵔ ⴷ ⴰⵢⵜ ⵜⵉⵢⵓⵔⵉ, ⵓⵍⴰ ⵉⵎⵓⴹⴹⴰⵕ."""
    example_3 = "ⴷⴰⴳ ⵓⵢⵍⵅ ⵙ ⵉⴳⵏⵏⴰ|ⵏⵏⴰⵏ ⵉⵢⵉ|ⴳⴳⵯⵣ ⴷ!|ⵏⵏⵉⵅ ⴰⵙⵏ|ⵜⵎⵢⴰⵔⵎ ⴰⵣⴷⴷⵉⵔ|ⵜⵎⵢⴰⵔⵎ ⵜⴰⵍⵍⴰ ⴷ ⵉⵎⵟⵟⴰⵡⵏ"
    examples = gr.Examples(
        examples=[
            ["common_voice_zgh_37837257.mp3", None, "ⵎⵍ ⵉⵢⵉ ⵎⴰⴷ ⴷ ⵜⴻⵜⵜⵎⵓⵏⴷ ⴰⴷ ⴰⴽ ⵎⵍⵖ ⵎⴰⴷ ⵜⴳⵉⴷ"],
            ["Voice1410.wav", None, example_2],
            ["Tamazight_For_All.mp3", None, example_3]
        ],
        inputs=[audio_file_in, subs_file_in, ref_text]
    )
    demo.queue()
    demo.launch()