"import numpy, scipy, matplotlib.pyplot as plt, librosa, urllib, IPython.display"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"[← Back to Index](index.html)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Exercise: Source Separation using NMF"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Goals:\n",
"\n",
"1. Separate sources using NMF.\n",
"2. Analyze and classify separated sources."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Download an audio file:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"urllib.urlretrieve?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Load an audio file:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"librosa.load?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Plot the spectrogram:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"librosa.stft?"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"librosa.logamplitude?"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"librosa.display.specshow?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Use NMF to decompose the spectrogram:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"librosa.decompose.decompose?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Plot the spectral profiles and temporal activations of each component individually:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"plt.subplot?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Use the inverse STFT to synthesize the separated sources. (`librosa.istft` expects a full, complex-valued spectrogram. Therefore, you will need to include the phase of the original spectrogram in your reconstruction. See [the notebook on NMF](nmf.html) for more details.)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"librosa.istft?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Use the reconstructed outputs as inputs into the kick/snare classifier that you created in an earlier exercise. Observe the results; are you able to automatically classify the separated sources?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## For Further Exploration"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Use different audio files."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Alter the rank of the decomposition, `n_components`. What happens when `n_components` is too large? too small?"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"NMF is a useful preprocessor for MIR tasks such as music transcription. Using the steps above, build your own simple transcription system that returns a sequence of note events, `[(onset time, class label, volume/gain)...]`."
