import time import streamlit as st # from transformers import pipeline import os import torch import datetime import numpy as np import soundfile from wavmark.utils import file_reader from audioseal import AudioSeal import torchaudio from pydub import AudioSegment import io # pipeline = pipeline(task="image-classification", model="julien-c/hotdog-not-hotdog") # st.title("Hot Dog? Or Not?") # file_name = st.file_uploader("Upload a hot dog candidate image") # if file_name is not None: # col1, col2 = st.columns(2) # image = Image.open(file_name) # col1.image(image, use_column_width=True) # predictions = pipeline(image) # col2.header("Probabilities") # for p in predictions: # col2.subheader(f"{ p['label'] }: { round(p['score'] * 100, 1)}%") # def read_as_single_channel_16k(audio_file, def_sr=16000, verbose=True, aim_second=None): # assert os.path.exists(audio_file) # st.markdown(os.path.exists(audio_file)) # file_extension = os.path.splitext(audio_file)[1].lower() # st.markdown(file_extension) # if file_extension == ".mp3": # data, origin_sr = librosa.load(audio_file, sr=None) # elif file_extension in [".wav", ".flac"]: # data, origin_sr = soundfile.read(audio_file) # else: # raise Exception("unsupported file:" + file_extension) # # channel check # if len(data.shape) == 2: # left_channel = data[:, 0] # if verbose: # print("Warning! the input audio has multiple chanel, this tool only use the first channel!") # data = left_channel # # sample rate check # if origin_sr != def_sr: # data = resampy.resample(data, origin_sr, def_sr) # if verbose: # print("Warning! The original samplerate is not 16Khz; the watermarked audio will be re-sampled to 16KHz") # sr = def_sr # audio_length_second = 1.0 * len(data) / sr # # if verbose: # # print("input length :%d second" % audio_length_second) # if aim_second is not None: # signal = data # assert len(signal) > 0 # current_second = len(signal) / sr # if current_second < aim_second: # repeat_count = int(aim_second / current_second) + 1 # signal = np.repeat(signal, repeat_count) # data = signal[0:sr * aim_second] # return data, sr, audio_length_second # def my_read_file(audio_path, max_second): # signal, sr, audio_length_second = read_as_single_channel_16k(audio_path, default_sr) # if audio_length_second > max_second: # signal = signal[0:default_sr * max_second] # audio_length_second = max_second # return signal, sr, audio_length_second def create_default_value(): if "def_value" not in st.session_state: def_val_npy = np.random.choice([0, 1], size=32 - len_start_bit) def_val_str = "".join([str(i) for i in def_val_npy]) st.session_state.def_value = def_val_str def download_sample_audio(): url = "https://keithito.com/LJ-Speech-Dataset/LJ037-0171.wav" with open("test.wav", "wb") as f: resp = urllib.request.urlopen(url) f.write(resp.read()) wav, sample_rate = torchaudio.load("test.wav") return wav, sample_rate # Main web app def main(): create_default_value() # st.title("MDS07") # st.write("https://github.com/wavmark/wavmark") markdown_text = """ # MDS07 [AudioSeal](https://github.com/jcha0155/AudioSealEnhanced) is the next-generation watermarking tool driven by AI. You can upload an audio file and encode a custom 16-bit watermark or perform decoding from a watermarked audio. This page is for demonstration usage and only process **the first minute** of the audio. If you have longer files for processing, we recommend using [our python toolkit](https://github.com/jcha0155/AudioSealEnhanced). """ # 使用st.markdown渲染Markdown文本 st.markdown(markdown_text) audio_file = st.file_uploader("Upload Audio", type=["wav", "mp3"], accept_multiple_files=False) if audio_file: # 保存文件到本地： # tmp_input_audio_file = os.path.join("/tmp/", audio_file.name) # st.markdown(tmp_input_audio_file) # with open(tmp_input_audio_file, "wb") as f: # f.write(audio_file.getbuffer()) # st.audio(tmp_input_audio_file, format="mp3/wav") # Save file to local storage tmp_input_audio_file = os.path.join("/tmp/", audio_file.name) with open("test.wav", "wb") as f: f.write(audio_file.getbuffer()) # # Convert MP3 to WAV using pydub # mp3_audio = AudioSegment.from_mp3(tmp_input_audio_file) # wav_output_file = tmp_input_audio_file.replace(".mp3", ".wav") # mp3_audio.export(wav_output_file, format="wav") # Load the WAV file using torchaudio wav, sample_rate = torchaudio.load("test.wav") st.markdown("Before unsquuezewav") st.markdown(wav) wav= wav.unsqueeze(0) #2nd way # Convert the tensor to a byte-like object in WAV format with io.BytesIO() as buffer: # Save the audio to the buffer using torchaudio torchaudio.save(buffer, wav, default_sr, format="wav") # Get the byte data from the buffer wav = buffer.getvalue() # Play the audio file (WAV format) st.audio(wav, format="audio/wav") # wav, sample_rate = torchaudio.load(audio_file, format="mp3/wav") st.markdown("SR") st.markdown(sample_rate) st.markdown("after unsqueeze wav") st.markdown(wav) # 展示文件到页面上 # st.audio(tmp_input_audio_file, format="audio/wav") action = st.selectbox("Select Action", ["Add Watermark", "Decode Watermark"]) if action == "Add Watermark": watermark_text = st.text_input("The watermark (0, 1 list of length-16):", value=st.session_state.def_value) add_watermark_button = st.button("Add Watermark", key="add_watermark_btn") if add_watermark_button: # 点击按钮后执行的 if audio_file and watermark_text: with st.spinner("Adding Watermark..."): #wav = my_read_file(wav,max_second_encode) #1st attempt watermark = model.get_watermark(wav, default_sr) watermarked_audio = wav + watermark print(watermarked_audio.size()) size = watermarked_audio.size() st.markdown(size) print(watermarked_audio.squeeze()) squeeze = watermarked_audio.squeeze(1) shape = squeeze.size() st.markdown(shape) st.markdown(squeeze) watermarked_audio = torchaudio.save("output.wav", squeeze, default_sr) st.audio(watermarked_audio, format="audio/wav") #2nd Attempt # watermarked_audio = model(wav, sample_rate=default_sr, alpha=1) # print(watermarked_audio.size()) # size = watermarked_audio.size() # st.markdown(size) # print(watermarked_audio.squeeze()) # squeeze = watermarked_audio.squeeze(1) # shape = squeeze.size() # st.markdown(shape) # st.markdown(squeeze) # # watermarked_audio, encode_time_cost = add_watermark(tmp_input_audio_file, watermark_text) # st.write("Watermarked Audio:") # st.markdown(watermarked_audio) # print("watermarked_audio:", watermarked_audio) # watermarked_audio = torchaudio.save("output.wav", squeeze, default_sr) # st.audio(watermarked_audio, format="audio/wav") #st.write("Time Cost: %d seconds" % encode_time_cost) # # st.button("Add Watermark", disabled=False) # elif action == "Decode Watermark": # if st.button("Decode"): # with st.spinner("Decoding..."): # decode_watermark(tmp_input_audio_file) if __name__ == "__main__": default_sr = 16000 max_second_encode = 60 max_second_decode = 30 len_start_bit = 16 device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') # model = wavmark.load_model().to(device) model = AudioSeal.load_generator("audioseal_wm_16bits") main() # audio_path = "/Users/my/Library/Mobile Documents/com~apple~CloudDocs/CODE/PycharmProjects/4_语音水印/419_huggingface水印/WavMark/example.wav" # decoded_watermark, decode_cost = decode_watermark(audio_path) # print(decoded_watermark)