File size: 13,604 Bytes
eb9c81a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
import typing as tp
from dataclasses import dataclass


import torch
import math
import numpy as np
import torch.nn as nn
from torch import Tensor
import torchaudio.functional as F
from torchaudio.transforms import MelScale

from .streaming import StreamingModule
from .conv import get_extra_padding_for_conv1d, pad1d


class LinearSpectrogram(nn.Module):
    def __init__(self, n_fft=1024, win_length=1024, hop_length=320, mode="pow2_sqrt"):
        """
        Initializes the streaming spectrogram module.
        
        Parameters:
            n_fft (int): Number of FFT points.
            win_length (int): Window length (in samples).
            hop_length (int): Hop length (in samples).
            mode (str): Calculation mode. "pow2_sqrt" computes magnitude as sqrt(sum(squared)).
        """
        super().__init__()
        self.n_fft = n_fft
        self.win_length = win_length
        self.hop_length = hop_length
        self.mode = mode
        
        # Register the Hann window as a buffer.
        self.register_buffer("window", torch.hann_window(win_length), persistent=False)
        

    def forward(self, y: torch.Tensor) -> torch.Tensor:
        """
        Computes the spectrogram from an input waveform chunk.
        
        Parameters:
            y (Tensor): Input waveform of shape (batch, time).
                                      
        Returns:
            spec (Tensor): Spectrogram tensor of shape (batch, n_frames, n_fft//2+1).
        """
        batch_size, n_samples = y.shape
        device = y.device

        
        total_length = y.size(1)
        # Compute the number of full frames available.
        n_frames = (total_length - self.win_length) // self.hop_length + 1

        if n_frames <= 0:
            # Not enough samples to form one full frame: update state and return an empty spectrogram.
            return torch.empty(batch_size, self.n_fft // 2 + 1, 0, device=device)

        
        # Determine the number of samples used in forming complete frames.
        used_length = (n_frames - 1) * self.hop_length + self.win_length

        # if used_length < total_length:
        #     warnings.warn(f"Extra {total_length - used_length} samples will be discarded to form complete frames.")
        
        y = y[:, :used_length]
        
        # Extract overlapping frames using unfolding.
        # The resulting shape is (batch, n_frames, win_length)
        frames = y.unfold(dimension=1, size=self.win_length, step=self.hop_length)
        
        # Apply the window to each frame.
        frames = frames * self.window
        
        # Compute the real FFT on each frame.
        spec = torch.fft.rfft(frames, n=self.n_fft)
        
        # If using "pow2_sqrt" mode, compute the magnitude spectrogram.
        if self.mode == "pow2_sqrt":
            # Compute sqrt(real^2 + imag^2) with epsilon for numerical stability.
            spec = torch.sqrt(spec.real.pow(2) + spec.imag.pow(2) + 1e-6)
        
        spec = spec.transpose(1, 2)  # (batch, n_fft//2+1, n_frames)
        return spec



class LogMelSpectrogram(nn.Module):
    def __init__(
        self,
        sample_rate=16000,
        n_fft=1024,
        win_length=1024,
        hop_length=320,
        n_mels=128,
        f_min=0.0,
        f_max=None,
    ):
        super().__init__()

        self.sample_rate = sample_rate
        self.n_fft = n_fft
        self.win_length = win_length
        self.hop_length = hop_length
        self.n_mels = n_mels
        self.f_min = f_min
        self.f_max = f_max or float(sample_rate // 2)

        self.spectrogram = LinearSpectrogram(n_fft, win_length, hop_length)

        fb = F.melscale_fbanks(
            n_freqs=self.n_fft // 2 + 1,
            f_min=self.f_min,
            f_max=self.f_max,
            n_mels=self.n_mels,
            sample_rate=self.sample_rate,
            norm="slaney",
            mel_scale="slaney",
        )
        self.register_buffer(
            "fb",
            fb,
            persistent=False,
        )

    def compress(self, x: Tensor) -> Tensor:
        return torch.log(torch.clamp(x, min=1e-5))

    def decompress(self, x: Tensor) -> Tensor:
        return torch.exp(x)

    def apply_mel_scale(self, x: Tensor) -> Tensor:
        return torch.matmul(x.transpose(-1, -2), self.fb).transpose(-1, -2)

    def forward(
        self, x: Tensor, return_linear: bool = False, sample_rate: int = None
    ) -> Tensor:
        x = x.squeeze(1)
        if sample_rate is not None and sample_rate != self.sample_rate:
            x = F.resample(x, orig_freq=sample_rate, new_freq=self.sample_rate)

        linear = self.spectrogram(x)

        if linear.shape[-1] != 0:
            mel = self.apply_mel_scale(linear)
            mel = self.compress(mel)
        else:
            # Not enough samples to form one full frame: return an empty spectrogram.
            mel = torch.empty(
                linear.shape[0], self.n_mels, 0, device=linear.device, dtype=linear.dtype
            )

        compressed_linear = self.compress(linear) if linear.shape[-1] != 0 else linear
        if return_linear:
            return mel, compressed_linear

        return mel


@dataclass
class _StreamingSpecState:
    previous: torch.Tensor | None = None

    def reset(self):
        self.previous = None


class RawStreamingLogMelSpectrogram(LogMelSpectrogram, StreamingModule[_StreamingSpecState]):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

        assert (
            self.hop_length <= self.win_length
        ), "stride must be less than kernel_size."

    def _init_streaming_state(self, batch_size: int) -> _StreamingSpecState:
        return _StreamingSpecState()

    def forward(self, input: torch.Tensor) -> torch.Tensor:
        stride = self.hop_length
        kernel = self.win_length
        if self._streaming_state is None:
            return super().forward(input)
        else:
            # Due to the potential overlap, we might have some cache of the previous time steps.
            previous = self._streaming_state.previous
            if previous is not None:
                input = torch.cat([previous, input], dim=-1)
            B, C, T = input.shape
            # We now compute the number of full convolution frames, i.e. the frames
            # that are ready to be computed.
            num_frames = max(0, int(math.floor((T - kernel) / stride) + 1))
            offset = num_frames * stride
            # We will compute `num_frames` outputs, and we are advancing by `stride`
            # for each of the frame, so we know the data before `stride * num_frames`
            # will never be used again.
            self._streaming_state.previous = input[..., offset:]
            if num_frames > 0:
                input_length = (num_frames - 1) * stride + kernel
                out = super().forward(input[..., :input_length])
            else:
                # Not enough data as this point to output some new frames.
                out = torch.empty(
                    B, self.n_mels, 0, device=input.device, dtype=input.dtype
                )
            return out



@dataclass
class _StreamingLogMelSpecState:
    padding_to_add: int
    original_padding_to_add: int

    def reset(self):
        self.padding_to_add = self.original_padding_to_add


class StreamingLogMelSpectrogram(StreamingModule[_StreamingLogMelSpecState]):
    """LogMelSpectrogram with some builtin handling of asymmetric or causal padding
    """

    def __init__(
        self,
        sample_rate: int = 16000,
        n_fft: int = 1024,
        win_length: int = 1024,
        hop_length: int = 320,
        n_mels: int = 128,
        f_min: float = 0.0,
        f_max: float = None,
        causal: bool = False,
        pad_mode: str = "reflect",
    ):
        super().__init__()

        self.conv = RawStreamingLogMelSpectrogram(
            sample_rate=sample_rate,
            n_fft=n_fft,
            win_length=win_length,
            hop_length=hop_length,
            n_mels=n_mels,
            f_min=f_min,
            f_max=f_max,
        )
        self.causal = causal
        self.pad_mode = pad_mode

    @property
    def _stride(self) -> int:
        return self.conv.hop_length

    @property
    def _kernel_size(self) -> int:
        return self.conv.win_length

    @property
    def _effective_kernel_size(self) -> int:
        return self._kernel_size

    @property
    def _padding_total(self) -> int:
        return self._effective_kernel_size - self._stride

    def _init_streaming_state(self, batch_size: int) -> _StreamingLogMelSpecState:
        assert self.causal, "streaming is only supported for causal convs"
        return _StreamingLogMelSpecState(self._padding_total, self._padding_total)

    def forward(self, x):
        B, C, T = x.shape
        padding_total = self._padding_total
        extra_padding = get_extra_padding_for_conv1d(
            x, self._effective_kernel_size, self._stride, padding_total
        )
        state = self._streaming_state
        if state is None:
            if self.causal:
                # Left padding for causal
                x = pad1d(x, (padding_total, extra_padding), mode=self.pad_mode)
            else:
                # Asymmetric padding required for odd strides
                padding_right = padding_total // 2
                padding_left = padding_total - padding_right
                x = pad1d(
                    x, (padding_left, padding_right + extra_padding), mode=self.pad_mode
                )
        else:
            if state.padding_to_add > 0 and x.shape[-1] > 0:
                x = pad1d(x, (state.padding_to_add, 0), mode=self.pad_mode)
                state.padding_to_add = 0
        return self.conv(x)


class OverlapAdd1d(nn.Module):
    """
    Fixed ConvTranspose1d that performs overlap-add:
      in_channels = win_length, out_channels = 1, kernel=win_length, stride=hop.
    """
    def __init__(self, win_length: int, hop: int):
        super().__init__()
        self.deconv = nn.ConvTranspose1d(
            in_channels=win_length,
            out_channels=1,
            kernel_size=win_length,
            stride=hop,
            bias=False,
        )
        # build identity‑impulse weights
        w = torch.zeros(win_length, 1, win_length)
        for c in range(win_length):
            w[c, 0, c] = 1.0
        self.deconv.weight.data.copy_(w)
        self.deconv.weight.requires_grad_(False)

    def forward(self, x: Tensor) -> Tensor:
        # x: (B, win_length, F) → y: (B, 1, (F-1)*hop + win_length)
        y = self.deconv(x)
        return y.squeeze(1)  # → (B, T_out)


@dataclass
class _StreamingISTFTState:
    prev_buffer: Tensor

    def reset(self):
        self.prev_buffer.zero_()


class StreamingISTFT(StreamingModule[_StreamingISTFTState]):
    """
    Streaming ISTFT via overlap-add:
      - inverse FFT per frame
      - window multiplication
      - overlap-add using a fixed ConvTranspose1d
      - carry tail for next chunk
    """
    def __init__(
        self,
        n_fft: int,
        hop_length: int,
        win_length: int | None = None,
    ):
        super().__init__()
        self.n_fft = n_fft
        self.hop = hop_length
        self.win_length = win_length or n_fft

        # hann window
        win = torch.hann_window(self.win_length)
        self.register_buffer("window", win, persistent=False)

        # overlap‑add helper
        self.overlap_add = OverlapAdd1d(self.win_length, self.hop)

        # how many samples to carry
        self.tail = self.win_length - self.hop
        if self.tail < 0:
            raise ValueError("hop_length must be <= win_length")

    def _init_streaming_state(self, batch_size: int) -> _StreamingISTFTState:
        buf = torch.zeros(batch_size, self.tail, device=self.window.device)
        return _StreamingISTFTState(prev_buffer=buf)

    def forward(self, S: Tensor) -> Tensor:
        """
        Args:
          S: complex STFT chunk, shape (B, n_fft//2+1, F_frames)
        Returns:
          time-domain chunk, shape (B, F_frames * hop_length)
        """
        B, n_freq, F = S.shape

        if F == 0:
            # Not enough samples to form one full frame: return an empty tensor.
            return torch.empty(B, 0, device=S.device, dtype=S.dtype)

        state = self._streaming_state
        # if state is None:
        #     # no streaming state, just do a regular ISTFT
        #     return torch.istft(
        #         S,
        #         n_fft=self.n_fft,
        #         hop_length=self.hop,
        #         win_length=self.win_length,
        #         window=self.window,
        #         center=False,
        #     )

        # (B, F, n_fft//2+1) -> real frames (B, F, n_fft)
        spec = S.transpose(1, 2)
        frames = torch.fft.irfft(spec, n=self.n_fft)
        # window + truncate
        frames = frames[..., : self.win_length] * self.window

        # (B, F, win) -> (B, win, F)
        x = frames.transpose(1, 2)

        # overlap-add in one CUDA kernel
        out = self.overlap_add(x)  # (B, F*hop + tail)

        if state is not None:
            # add carried‑over overlap
            out[:, : self.tail] += self._streaming_state.prev_buffer

            # slice off ready frames and save new tail
            ready = out[:, : F * self.hop]
            tail = out[:, F * self.hop :]
            self._streaming_state.prev_buffer = tail
        else:
            ready = out[:, : F * self.hop]
        return ready