Loading Chunked Data Error

[Ikerlandarech]

Hey there, currently loading NSynth dataset via DeepLake I'm trying to load a simple audio file audio = ds_train.audios[i][0] but I get this ReadSampleFromChunkError: Unable to read sample at index i from chunk 'audios/chunks/38c65994' in tensor audios.
If anyone can help on how to read chunked data into samples to work with them directly I would be so grateful.
Thank u so much in advance :)

[mikayelh]

HI @Ikerlandarech , thanks for flagging the issue. Looping in @fayaz-al who will take a look into it.

[Ikerlandarech]

Perfect thank you so much!

[fayaz-al]

Hi @Ikerlandarech! Can you please share the full error traceback? I am able to fetch samples from that chunk so I need to take a closer look.

[Ikerlandarech]

Sure thank you for helping me maybe is the problem is that I don't know how to load samples properly here's what I've done.

I have loaded my training dataset (Nsynth) directly from deeplake:
ds_train = deeplake.load("hub://activeloop/nsynth-train")
As we can see this is the summary of the dataset I thought to load an specific sample I should load them from the audios tensor.

As we can observe audio = ds_train.audios has shape (289205, 64000, 1) . audio = ds_train.audios[0] has therefore shape (64000, 1) which I think may correspond to all the datapoints of 1 of the 289205 samples of the dataset.

Now here's when I think I don't understand what is happening, what I wanted to do is to play that audio as: play_audio(audio, sample_rate)
Defining the function properly as:
def play_audio(audio, sample_rate): ipd.display(ipd.Audio(audio, rate=sample_rate))
The problem is that when running the function with audio = ds_train.audios[0] or audio = ds_train.audios[0][0] I get the same ReadSampleFromChunkError: Unable to read sample at index 0 from chunk 'audios/chunks/2820a70' in tensor audios.

I will share below the full error traceback, thank you again so very much for helping me out, I really appreciate it :)

`

ModuleNotFoundError Traceback (most recent call last)

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\compression.py:1036, in _open_audio(file)
1035 try:
-> 1036 import av
1037 except ImportError:

ModuleNotFoundError: No module named 'av'

During handling of the above exception, another exception occurred:

ModuleNotFoundError Traceback (most recent call last)
File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\compression.py:291, in decompress_array(buffer, shape, dtype, compression, start_idx, end_idx, step, reverse, to_pil, path)
290 elif compr_type == AUDIO_COMPRESSION:
--> 291 return _decompress_audio(buffer)
292 elif compr_type == VIDEO_COMPRESSION:

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\compression.py:1110, in _decompress_audio(file)
1109 def _decompress_audio(file: Union[bytes, memoryview, str]):
-> 1110 container, astream = _open_audio(file)
1111 shape = _read_shape_from_astream(container, astream)

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\compression.py:1038, in _open_audio(file)
1037 except ImportError:
-> 1038 raise ModuleNotFoundError(
1039 "PyAV is not installed. Please run pip install deeplake[audio]"
1040 )
1042 if isinstance(file, str):

ModuleNotFoundError: PyAV is not installed. Please run pip install deeplake[audio]

The above exception was the direct cause of the following exception:

SampleDecompressionError Traceback (most recent call last)
File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk\base_chunk.py:638, in catch_chunk_read_error..wrapper(self, *args, **kwargs)
637 try:
--> 638 return fn(self, *args, **kwargs)
639 except EmptyTensorError:

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk\sample_compressed_chunk.py:138, in SampleCompressedChunk.read_sample(self, local_index, cast, copy, sub_index, stream, decompress, is_tile, to_pil)
132 return Polygons.frombuffer(
133 bytes(buffer),
134 dtype=self.tensor_meta.dtype,
135 ndim=shape[-1],
136 )
--> 138 sample = decompress_array(
139 buffer,
140 shape,
141 self.dtype,
142 self.compression,
143 start_idx=start,
144 end_idx=stop,
145 step=step,
146 reverse=reverse,
147 to_pil=to_pil,
148 )
149 if to_pil:

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\compression.py:310, in decompress_array(buffer, shape, dtype, compression, start_idx, end_idx, step, reverse, to_pil, path)
309 except Exception as e:
--> 310 raise SampleDecompressionError(path) from e
312 try:

SampleDecompressionError: Could not decompress sample. Either the sample's buffer is corrupted, or it is in an unsupported format. Supported compressions: ['avi', 'bmp', 'dcm', 'dib', 'eps', 'flac', 'fli', 'gif', 'ico', 'im', 'jpeg', 'jpeg2000', 'las', 'lz4', 'mkv', 'mp3', 'mp4', 'mpo', 'msp', 'nii', 'nii.gz', 'pcx', 'ply', 'png', 'ppm', 'sgi', 'tga', 'tiff', 'wav', 'webp', 'wmf', 'xbm', None].

The above exception was the direct cause of the following exception:

ReadSampleFromChunkError Traceback (most recent call last)
File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk_engine.py:2249, in ChunkEngine._get_samples(self, chunk_id, row, idxs, index, is_polygon, aslist, pad_tensor)
2248 local_idx = self.translate_to_local_index(idx, row)
-> 2249 sample = self.read_basic_sample_from_chunk(
2250 chunk_id, local_idx, index
2251 )
2252 else:

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk_engine.py:1994, in ChunkEngine.read_basic_sample_from_chunk(self, chunk_id, local_sample_index, index, worst_case_header_size, is_tile, decompress)
1991 decompress = decompress or (
1992 isinstance(chunk, ChunkCompressedChunk) or len(index) > 1
1993 )
-> 1994 ret = chunk.read_sample(
1995 local_sample_index,
1996 cast=self.tensor_meta.htype != "dicom",
1997 is_tile=is_tile,
1998 decompress=decompress,
1999 )
2000 if len(index) > 1:

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk\base_chunk.py:642, in catch_chunk_read_error..wrapper(self, *args, **kwargs)
641 except Exception as e:
--> 642 raise ReadSampleFromChunkError(self.key) from e

ReadSampleFromChunkError: Unable to read sample from chunk 'audios/chunks/2820a70'.

The above exception was the direct cause of the following exception:

ReadSampleFromChunkError Traceback (most recent call last)
Cell In[130], line 4
2 audio.shape
3 # print(audio)
----> 4 play_audio(audio, sample_rate)

Cell In[7], line 31
30 def play_audio(audio, sample_rate):
---> 31 ipd.display(ipd.Audio(audio, rate=sample_rate))

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\IPython\lib\display.py:130, in Audio.init(self, data, filename, url, embed, rate, autoplay, normalize, element_id)
128 if rate is None:
129 raise ValueError("rate must be specified when data is a numpy array or list of audio samples.")
--> 130 self.data = Audio._make_wav(data, rate, normalize)

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\IPython\lib\display.py:152, in Audio._make_wav(data, rate, normalize)
149 import wave
151 try:
--> 152 scaled, nchan = Audio._validate_and_normalize_with_numpy(data, normalize)
153 except ImportError:
154 scaled, nchan = Audio._validate_and_normalize_without_numpy(data, normalize)

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\IPython\lib\display.py:172, in Audio._validate_and_normalize_with_numpy(data, normalize)
168 @staticmethod
169 def _validate_and_normalize_with_numpy(data, normalize) -> Tuple[bytes, int]:
170 import numpy as np
--> 172 data = np.array(data, dtype=float)
173 if len(data.shape) == 1:
174 nchan = 1

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\tensor.py:899, in Tensor.array(self, dtype)
898 def array(self, dtype=None) -> np.ndarray:
--> 899 ret = self.numpy() # type: ignore
900 if self.base_htype == "polygon":
901 return np.array(ret, dtype=dtype)

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\tensor.py:866, in Tensor.numpy(self, aslist, fetch_chunks)
841 def numpy(
842 self, aslist=False, fetch_chunks=False
843 ) -> Union[np.ndarray, List[np.ndarray]]:
844 """Computes the contents of the tensor in numpy format.
845
846 Args:
(...)
864 For tensors of htype polygon, aslist is always True.
865 """
--> 866 ret = self.chunk_engine.numpy(
867 self.index,
868 aslist=aslist,
869 fetch_chunks=fetch_chunks or self.is_iteration,
870 pad_tensor=self.pad_tensor,
871 )
872 if self.htype == "point_cloud": # TODO: refactor
873 if isinstance(ret, list):

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk_engine.py:1910, in ChunkEngine.numpy(self, index, aslist, use_data_cache, fetch_chunks, pad_tensor)
1908 self.check_link_ready()
1909 fetch_chunks = fetch_chunks or self._get_full_chunk(index)
-> 1910 return (self._sequence_numpy if self.is_sequence else self._numpy)(
1911 index, aslist, use_data_cache, fetch_chunks, pad_tensor
1912 )

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk_engine.py:2341, in ChunkEngine._numpy(self, index, aslist, use_data_cache, fetch_chunks, pad_tensor)
2335 samples = []
2336 if (
2337 fetch_chunks
2338 and not (self.tensor_meta.is_link or self.is_video)
2339 and not isinstance(self.base_storage, MemoryProvider)
2340 ):
-> 2341 samples = self.get_samples(index, aslist, pad_tensor)
2342 else:
2343 for global_sample_index in index.values[0].indices(length):

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk_engine.py:2289, in ChunkEngine.get_samples(self, index, aslist, pad_tensor)
2285 samples = {}
2286 for chunk_id, row, idxs, is_tile in self.load_chunks(
2287 list(index.values[0].indices(self.num_samples))
2288 ):
-> 2289 chunk_samples, chunk_last_shape = read_samples(chunk_id, row, idxs)
2290 check_sample_shape(chunk_last_shape, last_shape, self.key, index, aslist)
2291 samples.update(chunk_samples)

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\deeplake\core\chunk_engine.py:2257, in ChunkEngine._get_samples(self, chunk_id, row, idxs, index, is_polygon, aslist, pad_tensor)
2255 raise GetChunkError(e.chunk_key, idx, self.name) from e
2256 except ReadSampleFromChunkError as e:
-> 2257 raise ReadSampleFromChunkError(e.chunk_key, idx, self.name) from e
2258 check_sample_shape(sample.shape, last_shape, self.key, index, aslist)
2259 last_shape = sample.shape

ReadSampleFromChunkError: Unable to read sample at index 0 from chunk 'audios/chunks/2820a70' in tensor audios.

`

[Ikerlandarech]

Solved I just reinstalled deeplake[audio] and the problem is no longer there.

[FayazRahman]

@Ikerlandarech Fetching audio and video samples with deeplake requires the PyAV library. Install it using pip install av.

This is actually visible in the traceback but maybe we should do a better job of displaying it :)

[Ikerlandarech]

Hey there! Thank you so much for helping out, here's the problem I'm currently facing:
When trying to play an specific audio sample as:
audio = ds_train.audios[34][0]
play_audio(audio, sample_rate)
With:
def play_audio(audio, sample_rate): ipd.display(ipd.Audio(audio, rate=sample_rate))
The output I get is a 0s wav completely empty:

When doing the same operation defining:
audio = ds_train.audios[34]
I get the following error:
`

error Traceback (most recent call last)
Cell In[20], line 3
1 audio = ds_train.audios[34]
2 # print(audio)
----> 3 play_audio(audio, sample_rate)

Cell In[1], line 31
30 def play_audio(audio, sample_rate):
---> 31 ipd.display(ipd.Audio(audio, rate=sample_rate))

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\IPython\lib\display.py:130, in Audio.init(self, data, filename, url, embed, rate, autoplay, normalize, element_id)
128 if rate is None:
129 raise ValueError("rate must be specified when data is a numpy array or list of audio samples.")
--> 130 self.data = Audio._make_wav(data, rate, normalize)

File c:\Users\Usuario\Desktop\Academy Stuff\4t\TFG\INSTALLATIONS\TFG-DDSP\lib\site-packages\IPython\lib\display.py:162, in Audio._make_wav(data, rate, normalize)
160 waveobj.setsampwidth(2)
161 waveobj.setcomptype('NONE','NONE')
--> 162 waveobj.writeframes(scaled)
163 val = fp.getvalue()
164 waveobj.close()

File ~\AppData\Local\Programs\Python\Python310\lib\wave.py:437, in Wave_write.writeframes(self, data)
436 def writeframes(self, data):
--> 437 self.writeframesraw(data)
438 if self._datalength != self._datawritten:
439 self._patchheader()

File ~\AppData\Local\Programs\Python\Python310\lib\wave.py:426, in Wave_write.writeframesraw(self, data)
424 if not isinstance(data, (bytes, bytearray)):
425 data = memoryview(data).cast('B')
--> 426 self._ensure_header_written(len(data))
427 nframes = len(data) // (self._sampwidth * self._nchannels)
428 if self._convert:

File ~\AppData\Local\Programs\Python\Python310\lib\wave.py:467, in Wave_write._ensure_header_written(self, datasize)
465 if not self._framerate:
466 raise Error('sampling rate not specified')
--> 467 self._write_header(datasize)

File ~\AppData\Local\Programs\Python\Python310\lib\wave.py:479, in Wave_write._write_header(self, initlength)
477 except (AttributeError, OSError):
478 self._form_length_pos = None
--> 479 self._file.write(struct.pack('<L4s4sLHHLLHH4s',
480 36 + self._datalength, b'WAVE', b'fmt ', 16,
481 WAVE_FORMAT_PCM, self._nchannels, self._framerate,
482 self._nchannels * self._framerate * self._sampwidth,
483 self._nchannels * self._sampwidth,
484 self._sampwidth * 8, b'data'))
485 if self._form_length_pos is not None:
486 self._data_length_pos = self._file.tell()

error: ushort format requires 0 <= number <= 0xffff
`

If you could help me out on guiding me what's the correct way to load a certain sample (defining my audio variable) I would really appreciate it.

[FayazRahman]

ds.audio[34] would mean the audio sample at index 34 andds.audio[34][0] would mean the first frame of that sample.

So what you want is probably ds.audio[34]. But remember that this is still a Tensor object.

If you want a numpy array from this, you should do ds.audio[34].numpy().

[Ikerlandarech]

Hey there! Thank you so much everything working I found the problem. When loading for instance sample [25] as audio = ds_train.audios[25].numpy() it has shape (64000, 1) when it should be of type (1, 64000) so I was missing 2 lines:

audio = ds_train.audios[25].numpy()
audio = audio.reshape(-1)
audio = audio[np.newaxis, :]
play_audio(audio, sample_rate)

Now everything is working fine, thank you so much for helping me out!

[FayazRahman]

No problem, glad you got it to work!