![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1565256%2Fdec7edaf1ada541139d57c35e7d0672c%2Fkaggle.png?generation=1593110024594644&alt=media\")
This is my first audio classification competition and I am trying to convert librosa melspectrogram to image array (3D)
\n\ndef extract_features(file_name, sample_rate = 44100, as_mono = True , preemphasis = 0.95):\n\n sig, rate = librosa.load(file_name, sr = sample_rate, res_type='kaiser_fast') \n\n SHAPE = (128, 256) #(heigt, width) in pixels\n\n #Mel-spectrogram\n SAMPLE_RATE = rate\n N_FFT = SHAPE[0] * 8 # window length \n N_MELS = SHAPE[0]\n HOP_LEN = len(sig) // (SHAPE[1] - 1)\n #print(f'hop len = {HOP_LEN}')\n FMIN = 500\n FMAX = 15000\n # Preemphasis as in python_speech_features by James Lyons\n if preemphasis:\n sig = np.append(sig[0], sig[1:] - preemphasis * sig[:-1])\n #Librosa mel-spectrum\n melspec = librosa.feature.melspectrogram(y=sig, sr=SAMPLE_RATE, \n hop_length = HOP_LEN,\n n_mels = N_MELS,\n fmax = FMAX, fmin = FMIN,\n power = 1.0\n )\n\n #Convert power spec to db Scale\n melspec = librosa.amplitude_to_db(melspec, ref=np.max , top_db = 80)\n\n\n #Trim to desired shape \n melspec = melspec[:SHAPE[0], :SHAPE[1]]\n\n\n return melspec.astype('float32')` \nthe above function returns shape(128 , 256), when I try convert to 3D image array using:\nfrom PIL import Image
\n mel = extract_features('../input/birdsong-recognition/train_audio/aldfly/XC134874.mp3')\n mel = Image.fromarray(mel)
It returns blank image:
\n\n
Edit: Got it working. Read this kaggle Notebook: \n```\ndef mono_to_color(X, mean=None, std=None, norm_max=None, norm_min=None, eps=1e-6):\n # Stack X as [X,X,X]\n X = np.stack([X, X, X], axis=-1)
\n\n# Standardize\nmean = mean or X.mean()\nstd = std or X.std()\nXstd = (X - mean) / (std + eps)\n_min, _max = Xstd.min(), Xstd.max()\nnorm_max = norm_max or _max\nnorm_min = norm_min or _min\nif (_max - _min) > eps:\n # Scale to [0, 255]\n V = Xstd\n V[V < norm_min] = norm_min\n V[V > norm_max] = norm_max\n V = 255 * (V - norm_min) / (norm_max - norm_min)\n V = V.astype(np.uint8)\nelse:\n # Just zero\n V = np.zeros_like(Xstd, dtype=np.uint8)\nreturn V\n```\nResult:\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F1565256%2Fbd474e5261ff59e21f063fd548b8ae23%2FAnnotation%202020-07-02%20142202.png?generation=1593680019098131&alt=media\")
Do this : \nmel_spec = librosa.feature.melspectrogram(y=audio, sr=sr, n_mels=n_mels, hop_length=hop_length, fmin=fmin, fmax=fmax)
\n\nmel_db = librosa.power_to_db(mel_spec)
", "rawMarkdown": "Do this : \nmel\\_spec = librosa.feature.melspectrogram(y=audio, sr=sr, n\\_mels=n\\_mels, hop\\_length=hop\\_length, fmin=fmin, fmax=fmax)\n\nmel\\_db = librosa.**power\\_to\\_db**(mel\\_spec)", "replies": [ { "id": 912126, "postDate": "2020-07-02T08:44:51.653Z", "content": "Thanks for replying, but got the same result.
", "rawMarkdown": "Thanks for replying, but got the same result." } ] }, { "id": 901852, "postDate": "2020-06-25T18:37:52.057Z", "content": "This is my first audio classification competition and I am trying to convert librosa melspectrogram to image array (3D)
\n\ndef extract_features(file_name, sample_rate = 44100, as_mono = True , preemphasis = 0.95):\n\n sig, rate = librosa.load(file_name, sr = sample_rate, res_type='kaiser_fast') \n\n SHAPE = (128, 256) #(heigt, width) in pixels\n\n #Mel-spectrogram\n SAMPLE_RATE = rate\n N_FFT = SHAPE[0] * 8 # window length \n N_MELS = SHAPE[0]\n HOP_LEN = len(sig) // (SHAPE[1] - 1)\n #print(f'hop len = {HOP_LEN}')\n FMIN = 500\n FMAX = 15000\n # Preemphasis as in python_speech_features by James Lyons\n if preemphasis:\n sig = np.append(sig[0], sig[1:] - preemphasis * sig[:-1])\n #Librosa mel-spectrum\n melspec = librosa.feature.melspectrogram(y=sig, sr=SAMPLE_RATE, \n hop_length = HOP_LEN,\n n_mels = N_MELS,\n fmax = FMAX, fmin = FMIN,\n power = 1.0\n )\n\n #Convert power spec to db Scale\n melspec = librosa.amplitude_to_db(melspec, ref=np.max , top_db = 80)\n\n\n #Trim to desired shape \n melspec = melspec[:SHAPE[0], :SHAPE[1]]\n\n\n return melspec.astype('float32')` \nthe above function returns shape(128 , 256), when I try convert to 3D image array using:\nfrom PIL import Image
\n mel = extract_features('../input/birdsong-recognition/train_audio/aldfly/XC134874.mp3')\n mel = Image.fromarray(mel)
It returns blank image:
\n\n
Edit: Got it working. Read this kaggle Notebook: \n```\ndef mono_to_color(X, mean=None, std=None, norm_max=None, norm_min=None, eps=1e-6):\n # Stack X as [X,X,X]\n X = np.stack([X, X, X], axis=-1)
\n\n# Standardize\nmean = mean or X.mean()\nstd = std or X.std()\nXstd = (X - mean) / (std + eps)\n_min, _max = Xstd.min(), Xstd.max()\nnorm_max = norm_max or _max\nnorm_min = norm_min or _min\nif (_max - _min) > eps:\n # Scale to [0, 255]\n V = Xstd\n V[V < norm_min] = norm_min\n V[V > norm_max] = norm_max\n V = 255 * (V - norm_min) / (norm_max - norm_min)\n V = V.astype(np.uint8)\nelse:\n # Just zero\n V = np.zeros_like(Xstd, dtype=np.uint8)\nreturn V\n```\nResult:\n
Maybe all values in mel are 0?\nTry a quick diagnostic:\nplt.plot(mel.flatten()))\nYou should see the actual values as a line plot
", "rawMarkdown": "Maybe all values in mel are 0?\nTry a quick diagnostic:\nplt.plot(mel.flatten()))\nYou should see the actual values as a line plot", "isDeleted": true, "replies": [ { "id": 912127, "postDate": "2020-07-02T08:44:55.943Z", "content": "Yes, all mel values are below 0.
", "rawMarkdown": "Yes, all mel values are below 0. " } ] } ], "comments": [ { "id": 904627, "author_name": "CVxTz", "author_url": "", "post_date": "2020-06-27T19:17:25.367000", "content": "Do this : \nmel_spec = librosa.feature.melspectrogram(y=audio, sr=sr, n_mels=n_mels, hop_length=hop_length, fmin=fmin, fmax=fmax)
\n\nmel_db = librosa.power_to_db(mel_spec)
", "votes": 0, "replies": [ { "id": 912126, "author_name": "Neo", "author_url": "", "post_date": "2020-07-02T08:44:51.653000", "content": "Thanks for replying, but got the same result.
", "votes": 0, "replies": [] } ] }, { "id": 904212, "author_name": "", "author_url": "", "post_date": "2020-06-27T12:48:54.620000", "content": "Maybe all values in mel are 0?\nTry a quick diagnostic:\nplt.plot(mel.flatten()))\nYou should see the actual values as a line plot
", "votes": 0, "replies": [ { "id": 912127, "author_name": "Neo", "author_url": "", "post_date": "2020-07-02T08:44:55.943000", "content": "Yes, all mel values are below 0.
", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "904627": "Do this : \nmel\\_spec = librosa.feature.melspectrogram(y=audio, sr=sr, n\\_mels=n\\_mels, hop\\_length=hop\\_length, fmin=fmin, fmax=fmax)\n\nmel\\_db = librosa.**power\\_to\\_db**(mel\\_spec)", "901852": "This is my first audio classification competition and I am trying to convert librosa melspectrogram to image array (3D) \n \n def extract_features(file_name, sample_rate = 44100, as_mono = True , preemphasis = 0.95):\n \n sig, rate = librosa.load(file_name, sr = sample_rate, res_type='kaiser_fast') \n \n SHAPE = (128, 256) #(heigt, width) in pixels\n\n #Mel-spectrogram\n SAMPLE_RATE = rate\n N_FFT = SHAPE[0] * 8 # window length \n N_MELS = SHAPE[0]\n HOP_LEN = len(sig) // (SHAPE[1] - 1)\n #print(f'hop len = {HOP_LEN}')\n FMIN = 500\n FMAX = 15000\n # Preemphasis as in python_speech_features by James Lyons\n if preemphasis:\n sig = np.append(sig[0], sig[1:] - preemphasis * sig[:-1])\n #Librosa mel-spectrum\n melspec = librosa.feature.melspectrogram(y=sig, sr=SAMPLE_RATE, \n hop_length = HOP_LEN,\n n_mels = N_MELS,\n fmax = FMAX, fmin = FMIN,\n power = 1.0\n )\n \n #Convert power spec to db Scale\n melspec = librosa.amplitude_to_db(melspec, ref=np.max , top_db = 80)\n \n \n #Trim to desired shape \n melspec = melspec[:SHAPE[0], :SHAPE[1]]\n \n \n return melspec.astype('float32')` \n\nthe above function returns shape(128 , 256), when I try convert to 3D image array using:\n` from PIL import Image \n mel = extract_features('../input/birdsong-recognition/train_audio/aldfly/XC134874.mp3')\n mel = Image.fromarray(mel)`\n\nIt returns blank image:\n\n\n \n\n\nEdit: Got it working. Read this kaggle [Notebook](https://www.kaggle.com/daisukelab/cnn-2d-basic-solution-powered-by-fast-ai): \n```\ndef mono_to_color(X, mean=None, std=None, norm_max=None, norm_min=None, eps=1e-6):\n # Stack X as [X,X,X]\n X = np.stack([X, X, X], axis=-1)\n\n # Standardize\n mean = mean or X.mean()\n std = std or X.std()\n Xstd = (X - mean) / (std + eps)\n _min, _max = Xstd.min(), Xstd.max()\n norm_max = norm_max or _max\n norm_min = norm_min or _min\n if (_max - _min) > eps:\n # Scale to [0, 255]\n V = Xstd\n V[V < norm_min] = norm_min\n V[V > norm_max] = norm_max\n V = 255 * (V - norm_min) / (norm_max - norm_min)\n V = V.astype(np.uint8)\n else:\n # Just zero\n V = np.zeros_like(Xstd, dtype=np.uint8)\n return V\n```\nResult:\n\n", "904212": "Maybe all values in mel are 0?\nTry a quick diagnostic:\nplt.plot(mel.flatten()))\nYou should see the actual values as a line plot" } }