File size: 21,252 Bytes
bd6e54b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
import gradio as gr
import librosa
import numpy as np
import torch

from model.DiffSynthSampler import DiffSynthSampler
from tools import pad_STFT, encode_stft
from tools import safe_int, adjust_audio_length
from webUI.natural_language_guided_4.utils import InputBatch2Encode_STFT, encodeBatch2GradioOutput_STFT, \
    latent_representation_to_Gradio_image, resize_image_to_aspect_ratio, add_instrument


def get_sound2sound_with_text_module(gradioWebUI, sound2sound_with_text_state, virtual_instruments_state):
    # Load configurations
    uNet = gradioWebUI.uNet
    freq_resolution, time_resolution = gradioWebUI.freq_resolution, gradioWebUI.time_resolution
    VAE_scale = gradioWebUI.VAE_scale
    height, width, channels = int(freq_resolution / VAE_scale), int(time_resolution / VAE_scale), gradioWebUI.channels
    timesteps = gradioWebUI.timesteps
    VAE_encoder = gradioWebUI.VAE_encoder
    VAE_quantizer = gradioWebUI.VAE_quantizer
    VAE_decoder = gradioWebUI.VAE_decoder
    CLAP = gradioWebUI.CLAP
    CLAP_tokenizer = gradioWebUI.CLAP_tokenizer
    device = gradioWebUI.device
    squared = gradioWebUI.squared
    sample_rate = gradioWebUI.sample_rate
    noise_strategy = gradioWebUI.noise_strategy

    def receive_upload_origin_audio(sound2sound_duration, sound2sound_origin,

                                    sound2sound_with_text_dict, virtual_instruments_dict):
        origin_sr, origin_audio = sound2sound_origin
        origin_audio = origin_audio / np.max(np.abs(origin_audio))

        width = int(time_resolution * ((sound2sound_duration + 1) / 4) / VAE_scale)
        audio_length = 256 * (VAE_scale * width - 1)
        origin_audio = adjust_audio_length(origin_audio, audio_length, origin_sr, sample_rate)

        D = librosa.stft(origin_audio, n_fft=1024, hop_length=256, win_length=1024)
        padded_D = pad_STFT(D)
        encoded_D = encode_stft(padded_D)

        # Todo: justify batchsize to 1
        origin_spectrogram_batch_tensor = torch.from_numpy(
            np.repeat(encoded_D[np.newaxis, :, :, :], 1, axis=0)).float().to(device)

        # Todo: remove hard-coding
        origin_flipped_log_spectrums, origin_flipped_phases, origin_signals, origin_latent_representations, quantized_origin_latent_representations = InputBatch2Encode_STFT(
            VAE_encoder, origin_spectrogram_batch_tensor, resolution=(512, width * VAE_scale), quantizer=VAE_quantizer,
            squared=squared)

        sound2sound_with_text_dict["origin_latent_representations"] = origin_latent_representations.tolist()
        sound2sound_with_text_dict[
            "sound2sound_origin_latent_representation_image"] = latent_representation_to_Gradio_image(
            origin_latent_representations[0]).tolist()
        sound2sound_with_text_dict[
            "sound2sound_origin_quantized_latent_representation_image"] = latent_representation_to_Gradio_image(
            quantized_origin_latent_representations[0]).tolist()


        return {sound2sound_origin_spectrogram_image: resize_image_to_aspect_ratio(origin_flipped_log_spectrums[0],
                                                                                   1.55,
                                                                                   1),
                sound2sound_origin_phase_image: resize_image_to_aspect_ratio(origin_flipped_phases[0],
                                                                                   1.55,
                                                                                   1),
                sound2sound_origin_latent_representation_image: latent_representation_to_Gradio_image(
                    origin_latent_representations[0]),
                sound2sound_origin_quantized_latent_representation_image: latent_representation_to_Gradio_image(
                    quantized_origin_latent_representations[0]),
                sound2sound_with_text_state: sound2sound_with_text_dict,
                virtual_instruments_state: virtual_instruments_dict}

    def sound2sound_sample(sound2sound_prompts, sound2sound_negative_prompts, sound2sound_batchsize,

                           sound2sound_guidance_scale, sound2sound_sampler,

                           sound2sound_sample_steps,

                           sound2sound_noising_strength, sound2sound_seed, sound2sound_dict, virtual_instruments_dict):
        # input processing
        sound2sound_seed = safe_int(sound2sound_seed, 12345678)
        sound2sound_batchsize = int(sound2sound_batchsize)
        noising_strength = sound2sound_noising_strength
        sound2sound_sample_steps = int(sound2sound_sample_steps)
        CFG = int(sound2sound_guidance_scale)

        origin_latent_representations = torch.tensor(
            sound2sound_dict["origin_latent_representations"]).repeat(sound2sound_batchsize, 1, 1, 1).to(
            device)

        # sound2sound
        text2sound_embedding = \
            CLAP.get_text_features(**CLAP_tokenizer([sound2sound_prompts], padding=True, return_tensors="pt"))[0].to(
                device)

        mySampler = DiffSynthSampler(timesteps, height=height, channels=channels, noise_strategy=noise_strategy)
        negative_condition = \
            CLAP.get_text_features(**CLAP_tokenizer([sound2sound_negative_prompts], padding=True, return_tensors="pt"))[
                0]
        mySampler.activate_classifier_free_guidance(CFG, negative_condition.to(device))

        normalized_sample_steps = int(sound2sound_sample_steps / noising_strength)
        mySampler.respace(list(np.linspace(0, timesteps - 1, normalized_sample_steps, dtype=np.int32)))

        condition = text2sound_embedding.repeat(sound2sound_batchsize, 1)

        # Todo: remove-hard coding
        width = origin_latent_representations.shape[-1]
        new_sound_latent_representations, initial_noise = \
            mySampler.img_guided_sample(model=uNet, shape=(sound2sound_batchsize, channels, height, width),
                                        seed=sound2sound_seed,
                                        noising_strength=noising_strength,
                                        guide_img=origin_latent_representations, return_tensor=True,
                                        condition=condition,
                                        sampler=sound2sound_sampler)

        new_sound_latent_representations = new_sound_latent_representations[-1]

        # Quantize new sound latent representations
        quantized_new_sound_latent_representations, loss, (_, _, _) = VAE_quantizer(new_sound_latent_representations)

        new_sound_flipped_log_spectrums, new_sound_flipped_phases, new_sound_signals, _, _, _ = encodeBatch2GradioOutput_STFT(
            VAE_decoder,
            quantized_new_sound_latent_representations,
            resolution=(
                512,
                width * VAE_scale),
            original_STFT_batch=None
        )

        new_sound_latent_representation_gradio_images = []
        new_sound_quantized_latent_representation_gradio_images = []
        new_sound_spectrogram_gradio_images = []
        new_sound_phase_gradio_images = []
        new_sound_rec_signals_gradio = []
        for i in range(sound2sound_batchsize):
            new_sound_latent_representation_gradio_images.append(
                latent_representation_to_Gradio_image(new_sound_latent_representations[i]))
            new_sound_quantized_latent_representation_gradio_images.append(
                latent_representation_to_Gradio_image(quantized_new_sound_latent_representations[i]))
            new_sound_spectrogram_gradio_images.append(new_sound_flipped_log_spectrums[i])
            new_sound_phase_gradio_images.append(new_sound_flipped_phases[i])
            new_sound_rec_signals_gradio.append((sample_rate, new_sound_signals[i]))
        sound2sound_dict[
            "new_sound_latent_representation_gradio_images"] = new_sound_latent_representation_gradio_images
        sound2sound_dict[
            "new_sound_quantized_latent_representation_gradio_images"] = new_sound_quantized_latent_representation_gradio_images
        sound2sound_dict["new_sound_spectrogram_gradio_images"] = new_sound_spectrogram_gradio_images
        sound2sound_dict["new_sound_phase_gradio_images"] = new_sound_phase_gradio_images
        sound2sound_dict["new_sound_rec_signals_gradio"] = new_sound_rec_signals_gradio

        # save instrument
        sound2sound_dict["latent_representations"] = new_sound_latent_representations.to("cpu").detach().numpy()
        sound2sound_dict["quantized_latent_representations"] = quantized_new_sound_latent_representations.to(
            "cpu").detach().numpy()
        sound2sound_dict["condition"] = condition.to("cpu").detach().numpy()
        sound2sound_dict["negative_condition"] = negative_condition.to("cpu").detach().numpy()
        sound2sound_dict["guidance_scale"] = CFG
        sound2sound_dict["sampler"] = sound2sound_sampler

        return {sound2sound_new_sound_latent_representation_image: latent_representation_to_Gradio_image(
            new_sound_latent_representations[0]),
            sound2sound_new_sound_quantized_latent_representation_image: latent_representation_to_Gradio_image(
                quantized_new_sound_latent_representations[0]),
            sound2sound_new_sound_spectrogram_image: resize_image_to_aspect_ratio(new_sound_flipped_log_spectrums[0],
                                                                                   1.55,
                                                                                   1),
            sound2sound_new_sound_phase_image: resize_image_to_aspect_ratio(new_sound_flipped_phases[0],
                                                                                   1.55,
                                                                                   1),
            sound2sound_new_sound_audio: (sample_rate, new_sound_signals[0]),
            sound2sound_sample_index_slider: gr.update(minimum=0, maximum=sound2sound_batchsize - 1, value=0,
                                                       step=1.0,
                                                       visible=True,
                                                       label="Sample index",
                                                       info="Swipe to view other samples"),
            sound2sound_seed_textbox: sound2sound_seed,
            sound2sound_with_text_state: sound2sound_dict,
            virtual_instruments_state: virtual_instruments_dict}

    def show_sound2sound_sample(sound2sound_sample_index, sound2sound_with_text_dict):
        sample_index = int(sound2sound_sample_index)
        return {sound2sound_new_sound_latent_representation_image:
                    sound2sound_with_text_dict["new_sound_latent_representation_gradio_images"][sample_index],
                sound2sound_new_sound_quantized_latent_representation_image:
                    sound2sound_with_text_dict["new_sound_quantized_latent_representation_gradio_images"][sample_index],
                sound2sound_new_sound_spectrogram_image: resize_image_to_aspect_ratio(
                    sound2sound_with_text_dict["new_sound_spectrogram_gradio_images"][
                        sample_index], 1.55, 1),
                sound2sound_new_sound_phase_image: resize_image_to_aspect_ratio(
                    sound2sound_with_text_dict["new_sound_phase_gradio_images"][
                        sample_index], 1.55, 1),
                sound2sound_new_sound_audio: sound2sound_with_text_dict["new_sound_rec_signals_gradio"][sample_index]}

    def save_virtual_instrument(sample_index, virtual_instrument_name, sound2sound_dict, virtual_instruments_dict):
        virtual_instruments_dict = add_instrument(sound2sound_dict, virtual_instruments_dict, virtual_instrument_name,
                                                  sample_index)

        return {virtual_instruments_state: virtual_instruments_dict,
                text2sound_instrument_name_textbox: gr.Textbox(label="Instrument name", lines=1,
                                                               placeholder=f"Saved as {virtual_instrument_name}!")}

    with gr.Tab("Sound2Sound"):
        gr.Markdown("Generate new sound based on a given sound!")
        with gr.Row(variant="panel"):
            with gr.Column(scale=3):
                sound2sound_prompts_textbox = gr.Textbox(label="Positive prompt", lines=2, value="organ")
                text2sound_negative_prompts_textbox = gr.Textbox(label="Negative prompt", lines=2, value="")

            with gr.Column(scale=1):
                sound2sound_sample_button = gr.Button(variant="primary", value="Generate", scale=1)

                sound2sound_sample_index_slider = gr.Slider(minimum=0, maximum=3, value=0, step=1.0, visible=False,
                                                            label="Sample index",
                                                            info="Swipe to view other samples")

        with gr.Row(variant="panel"):
            with gr.Column(scale=1):
                with gr.Tab("Origin sound"):
                    sound2sound_duration_slider = gradioWebUI.get_duration_slider()

                    sound2sound_origin_audio = gr.Audio(
                        sources=["microphone", "upload"], label="Upload/Record source sound",
                        waveform_options=gr.WaveformOptions(
                            waveform_color="#01C6FF",
                            waveform_progress_color="#0066B4",
                            skip_length=1,
                            show_controls=False,
                        ),
                    )

                    with gr.Row(variant="panel"):
                        sound2sound_origin_spectrogram_image = gr.Image(label="Original upload spectrogram",
                                                                        type="numpy",visible=True)
                        sound2sound_origin_phase_image = gr.Image(label="Original upload phase",
                                                                  type="numpy", visible=True)

                with gr.Tab("Sound2sound settings"):
                    sound2sound_sample_steps_slider = gradioWebUI.get_sample_steps_slider()
                    sound2sound_sampler_radio = gradioWebUI.get_sampler_radio()
                    sound2sound_batchsize_slider = gradioWebUI.get_batchsize_slider()
                    sound2sound_noising_strength_slider = gradioWebUI.get_noising_strength_slider()
                    sound2sound_guidance_scale_slider = gradioWebUI.get_guidance_scale_slider()
                    sound2sound_seed_textbox = gradioWebUI.get_seed_textbox()

            with gr.Column(scale=1):
                sound2sound_new_sound_audio = gr.Audio(type="numpy", label="Play new sound", interactive=False,
                                                       waveform_options=gr.WaveformOptions(
                                                           waveform_color="#FFB6C1",
                                                           waveform_progress_color="#FF0000",
                                                           skip_length=1,
                                                           show_controls=False,
                                                       ), )
                with gr.Row(variant="panel"):
                    sound2sound_new_sound_spectrogram_image = gr.Image(label="New sound spectrogram", type="numpy",
                                                                       scale=1)
                    sound2sound_new_sound_phase_image = gr.Image(label="New sound phase", type="numpy",
                                                                 scale=1)

                with gr.Row(variant="panel",):
                    text2sound_instrument_name_textbox = gr.Textbox(label="Instrument name", lines=2,
                                                                    placeholder="Name of your instrument",
                                                                    scale=1)
                    text2sound_save_instrument_button = gr.Button(variant="primary",
                                                                  value="Save instrument",
                                                                  scale=1)

        with gr.Row(variant="panel"):
            sound2sound_origin_latent_representation_image = gr.Image(label="Original latent representation",
                                                                      type="numpy", height=800,
                                                                      visible=False)
            sound2sound_origin_quantized_latent_representation_image = gr.Image(
                label="Original quantized latent representation", type="numpy", height=800, visible=False)

            sound2sound_new_sound_latent_representation_image = gr.Image(label="New latent representation",
                                                                         type="numpy", height=800, visible=False)
            sound2sound_new_sound_quantized_latent_representation_image = gr.Image(
                label="New sound quantized latent representation", type="numpy", height=800, visible=False)

    sound2sound_origin_audio.change(receive_upload_origin_audio,
                                    inputs=[sound2sound_duration_slider,
                                            sound2sound_origin_audio,
                                            sound2sound_with_text_state,
                                            virtual_instruments_state],
                                    outputs=[sound2sound_origin_spectrogram_image,
                                             sound2sound_origin_phase_image,
                                             sound2sound_origin_latent_representation_image,
                                             sound2sound_origin_quantized_latent_representation_image,
                                             sound2sound_with_text_state,
                                             virtual_instruments_state])

    sound2sound_sample_button.click(sound2sound_sample,
                                    inputs=[sound2sound_prompts_textbox,
                                            text2sound_negative_prompts_textbox,
                                            sound2sound_batchsize_slider,
                                            sound2sound_guidance_scale_slider,
                                            sound2sound_sampler_radio,
                                            sound2sound_sample_steps_slider,
                                            sound2sound_noising_strength_slider,
                                            sound2sound_seed_textbox,
                                            sound2sound_with_text_state,
                                            virtual_instruments_state],
                                    outputs=[sound2sound_new_sound_latent_representation_image,
                                             sound2sound_new_sound_quantized_latent_representation_image,
                                             sound2sound_new_sound_spectrogram_image,
                                             sound2sound_new_sound_phase_image,
                                             sound2sound_new_sound_audio,
                                             sound2sound_sample_index_slider,
                                             sound2sound_seed_textbox,
                                             sound2sound_with_text_state,
                                             virtual_instruments_state])

    text2sound_save_instrument_button.click(save_virtual_instrument,
                                            inputs=[sound2sound_sample_index_slider,
                                                    text2sound_instrument_name_textbox,
                                                    sound2sound_with_text_state,
                                                    virtual_instruments_state],
                                            outputs=[virtual_instruments_state,
                                                     text2sound_instrument_name_textbox])

    sound2sound_sample_index_slider.change(show_sound2sound_sample,
                                           inputs=[sound2sound_sample_index_slider, sound2sound_with_text_state],
                                           outputs=[sound2sound_new_sound_latent_representation_image,
                                                    sound2sound_new_sound_quantized_latent_representation_image,
                                                    sound2sound_new_sound_spectrogram_image,
                                                    sound2sound_new_sound_phase_image,
                                                    sound2sound_new_sound_audio])