Module: TorchAudio

Defined in:
lib/torchaudio.rb,
lib/torchaudio/version.rb,
lib/torchaudio/functional.rb,
lib/torchaudio/datasets/utils.rb,
lib/torchaudio/datasets/yesno.rb,
lib/torchaudio/transforms/vol.rb,
lib/torchaudio/transforms/fade.rb,
lib/torchaudio/transforms/mfcc.rb,
lib/torchaudio/transforms/mel_scale.rb,
lib/torchaudio/transforms/spectrogram.rb,
lib/torchaudio/transforms/compute_deltas.rb,
lib/torchaudio/transforms/amplitude_to_db.rb,
lib/torchaudio/transforms/mel_spectrogram.rb,
lib/torchaudio/transforms/mu_law_decoding.rb,
lib/torchaudio/transforms/mu_law_encoding.rb

Defined Under Namespace

Modules: Datasets, Functional, Transforms Classes: Error

Constant Summary collapse

VERSION = 
"0.6.0"
F = 
Functional

Class Method Summary collapse

Class Method Details

.load(uri, frame_offset: 0, num_frames: -1,, channels_first: true) ⇒ Object 



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
 
# File 'lib/torchaudio.rb', line 30

def load(
  uri,
  frame_offset: 0,
  num_frames: -1,
  channels_first: true
)
  begin
    require "torchcodec"
  rescue LoadError
    raise LoadError, "TorchCodec is required for load. Please install torchcodec to use this function."
  end

  begin
    decoder = TorchCodec::Decoders::AudioDecoder.new(uri)
  rescue => e
    raise RuntimeError, "Failed to create AudioDecoder for #{uri}: #{e}"
  end

  # Get sample rate from metadata
  sample_rate = decoder.metadata[:sample_rate]
  if sample_rate.nil?
    raise RuntimeError, "Unable to determine sample rate from audio metadata"
  end

  # Decode the entire file first, then subsample manually
  # This is the simplest approach since torchcodec uses time-based indexing
  begin
    audio_samples = decoder.get_all_samples
  rescue => e
    raise RuntimeError, "Failed to decode audio samples: #{e}"
  end

  data = audio_samples[:data]

  # Apply frame_offset and num_frames (which are actually sample offsets)
  if frame_offset > 0
    if frame_offset >= data.shape[1]
      # Return empty tensor if offset is beyond available data
      empty_shape = channels_first ? [data.shape[0], 0] : [0, data.shape[0]]
      return [Torch.zeros(empty_shape, dtype: Torch.float32), sample_rate]
    end
    data = data[0.., frame_offset..]
  end

  if num_frames == 0
    # Return empty tensor if num_frames is 0
    empty_shape = channels_first ? [data.shape[0], 0] : [0, data.shape[0]]
    return [Torch.zeros(empty_shape, dtype: Torch.float32), sample_rate]
  elsif num_frames > 0
    data = data[0.., 0...num_frames]
  end

  # TorchCodec returns data in [channel, time] format by default
  # Handle channels_first parameter
  if !channels_first
    data = data.transpose(0, 1)  # [channel, time] -> [time, channel]
  end

  [data, sample_rate]
end

.load_wav(path, channels_first: true) ⇒ Object 



91
92
93
 
# File 'lib/torchaudio.rb', line 91

def load_wav(path, channels_first: true)
  load(path, channels_first: channels_first)
end

.save(uri, src, sample_rate, channels_first: true, compression: nil) ⇒ Object 



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
 
# File 'lib/torchaudio.rb', line 95

def save(
  uri,
  src,
  sample_rate,
  channels_first: true,
  compression: nil
)
  begin
    require "torchcodec"
  rescue LoadError
    raise LoadError, "TorchCodec is required for save. Please install torchcodec to use this function."
  end

  # Input validation
  if !src.is_a?(Torch::Tensor)
    raise ArgumentError, "Expected src to be a torch.Tensor, got #{src.class.name}"
  end

  if src.dtype != Torch.float32
    src = src.float
  end

  if sample_rate <= 0
    raise ArgumentError, "sample_rate must be positive, got #{sample_rate}"
  end

  # Handle tensor shape and channels_first
  if src.ndim == 1
    # Convert to 2D: [1, time] for channels_first: true
    if channels_first
      data = src.unsqueeze(0)  # [1, time]
    else
      # For channels_first: false, input is [time] -> reshape to [time, 1] -> transpose to [1, time]
      data = src.unsqueeze(1).transpose(0, 1)  # [time, 1] -> [1, time]
    end
  elsif src.ndim == 2
    if channels_first
      data = src  # Already [channel, time]
    else
      data = src.transpose(0, 1)  # [time, channel] -> [channel, time]
    end
  else
    raise ArgumentError, "Expected 1D or 2D tensor, got #{src.ndim}D tensor"
  end

  # Create AudioEncoder
  begin
    encoder = TorchCodec::Encoders::AudioEncoder.new(data, sample_rate: sample_rate)
  rescue => e
    raise RuntimeError, "Failed to create AudioEncoder: #{e}"
  end

  # Determine bit_rate from compression parameter
  bit_rate = nil
  if !compression.nil?
    if compression.is_a?(Integer) || compression.is_a?(Float)
      bit_rate = compression.to_i
    else
      warn "Unsupported compression type #{compression.class.name}."
    end
  end

  # Save to file
  begin
    encoder.to_file(uri, bit_rate: bit_rate)
  rescue => e
    raise RuntimeError, "Failed to save audio to #{uri}: #{e}"
  end
end