import soundfile as sf
import torch
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor,Wav2Vec2ProcessorWithLM
import gradio as gr
import sox
import subprocess


def read_file_and_process(wav_file):
    filename = wav_file.split('.')[0]
    filename_16k = filename + "16k.wav"
    resampler(wav_file, filename_16k)
    speech, _ = sf.read(filename_16k)
    inputs = processor(speech, sampling_rate=16_000, return_tensors="pt", padding=True)
    
    return inputs


def resampler(input_file_path, output_file_path):
    command = (
        f"ffmpeg -hide_banner -loglevel panic -i {input_file_path} -ar 16000 -ac 1 -bits_per_raw_sample 16 -vn "
        f"{output_file_path}"
    )
    subprocess.call(command, shell=True)


# def parse_transcription_with_lm(logits):
#     result = processor_with_LM.batch_decode(logits.cpu().numpy())
#     text = result.text
#     transcription = text[0].replace('<s>','')
#     return transcription

def parse_transcription(logits):
    predicted_ids = torch.argmax(logits, dim=-1)
    transcription = processor.decode(predicted_ids[0], skip_special_tokens=True)
    return transcription

# def parse(wav_file, applyLM):
#     input_values = read_file_and_process(wav_file)
#     with torch.no_grad():
#         logits = model(**input_values).logits
   
#     if applyLM:
#         # return parse_transcription_with_lm(logits)
#         return "done"
#     else:
#         return parse_transcription(logits)

def parse(wav_file, applyLM):
    input_values = read_file_and_process(wav_file)
    with torch.no_grad():
        logits = model(**input_values).logits

    if applyLM:
        # return parse_transcription_with_lm(logits)
        return "done"
    else:
        return parse_transcription(logits)

    
# model_id = "infinitejoy/wav2vec2-large-xls-r-300m-odia"
# working 50%  
# model_id = "Harveenchadha/odia_large_wav2vec2"

# It worked when first run but after that getting error
# model_id = "anuragshas/wav2vec2-large-xlsr-53-odia"

# model_id = "theainerd/wav2vec2-large-xlsr-53-odia"

# model_id = "Ranjit/Whisper_Small_Odia_CV_11.0_5k_steps"

# model_id = "theainerd/wav2vec2-large-xlsr-53-odia"

# model_id = "theainerd/wav2vec2-large-xlsr-53-odia"

# This is hindi
model_id = "Harveenchadha/vakyansh-wav2vec2-hindi-him-4200"



# processor = Wav2Vec2Processor.from_pretrained(model_id)
# # processor_with_LM = Wav2Vec2ProcessorWithLM.from_pretrained(model_id)
# model = Wav2Vec2ForCTC.from_pretrained(model_id)

    
# input_ = gr.Audio(source="microphone", type="filepath") 
# txtbox = gr.Textbox(
#             label="Output from model will appear here:",
#             lines=5
#         )
# chkbox = gr.Checkbox(label="Apply LM", value=False)


# gr.Interface(parse, inputs = [input_, chkbox],  outputs=txtbox,
#              streaming=True, interactive=True,
#              analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);





processor = Wav2Vec2Processor.from_pretrained(model_id)
model = Wav2Vec2ForCTC.from_pretrained(model_id)

# input_ = gr.inputs.File(source="upload", type="filepath")  # Change input source to "upload" and type to "audio"
input_ = gr.Audio(source="upload", type="filepath")
txtbox = gr.Textbox(
    label="Output from the model will appear here:",
    lines=5
)
chkbox = gr.Checkbox(label="Apply LM", value=False)

gr.Interface(parse, inputs=[input_, chkbox], outputs=txtbox,
             streaming=True, interactive=True,
             analytics_enabled=False, show_tips=False, enable_queue=True).launch(inline=False);