Spaces:
Runtime error
Runtime error
Commit
•
7834510
0
Parent(s):
Duplicate from mjaramillo/SpiceIcaroTP
Browse filesCo-authored-by: Miguel Jaramillo <mjaramillo@users.noreply.huggingface.co>
- .gitattributes +27 -0
- .github/workflows/main.yml +19 -0
- LICENSE +21 -0
- README.md +14 -0
- TP3.ipynb +0 -0
- app_deploy.py +506 -0
- requirements.txt +10 -0
.gitattributes
ADDED
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
*.7z filter=lfs diff=lfs merge=lfs -text
|
2 |
+
*.arrow filter=lfs diff=lfs merge=lfs -text
|
3 |
+
*.bin filter=lfs diff=lfs merge=lfs -text
|
4 |
+
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
5 |
+
*.ftz filter=lfs diff=lfs merge=lfs -text
|
6 |
+
*.gz filter=lfs diff=lfs merge=lfs -text
|
7 |
+
*.h5 filter=lfs diff=lfs merge=lfs -text
|
8 |
+
*.joblib filter=lfs diff=lfs merge=lfs -text
|
9 |
+
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
10 |
+
*.model filter=lfs diff=lfs merge=lfs -text
|
11 |
+
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
12 |
+
*.onnx filter=lfs diff=lfs merge=lfs -text
|
13 |
+
*.ot filter=lfs diff=lfs merge=lfs -text
|
14 |
+
*.parquet filter=lfs diff=lfs merge=lfs -text
|
15 |
+
*.pb filter=lfs diff=lfs merge=lfs -text
|
16 |
+
*.pt filter=lfs diff=lfs merge=lfs -text
|
17 |
+
*.pth filter=lfs diff=lfs merge=lfs -text
|
18 |
+
*.rar filter=lfs diff=lfs merge=lfs -text
|
19 |
+
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
20 |
+
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
21 |
+
*.tflite filter=lfs diff=lfs merge=lfs -text
|
22 |
+
*.tgz filter=lfs diff=lfs merge=lfs -text
|
23 |
+
*.wasm filter=lfs diff=lfs merge=lfs -text
|
24 |
+
*.xz filter=lfs diff=lfs merge=lfs -text
|
25 |
+
*.zip filter=lfs diff=lfs merge=lfs -text
|
26 |
+
*.zstandard filter=lfs diff=lfs merge=lfs -text
|
27 |
+
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
.github/workflows/main.yml
ADDED
@@ -0,0 +1,19 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
name: Sync to Hugging Face hub
|
2 |
+
on:
|
3 |
+
push:
|
4 |
+
branches: [main]
|
5 |
+
|
6 |
+
# to run this workflow manually from the Actions tab
|
7 |
+
workflow_dispatch:
|
8 |
+
|
9 |
+
jobs:
|
10 |
+
sync-to-hub:
|
11 |
+
runs-on: ubuntu-latest
|
12 |
+
steps:
|
13 |
+
- uses: actions/checkout@v2
|
14 |
+
with:
|
15 |
+
fetch-depth: 0
|
16 |
+
- name: Push to hub
|
17 |
+
env:
|
18 |
+
HF_TOKEN: ${{ secrets.HF_TOKEN }}
|
19 |
+
run: git push --force https://mjaramillo:$HF_TOKEN@huggingface.co/spaces/mjaramillo/SpiceIcaroTP
|
LICENSE
ADDED
@@ -0,0 +1,21 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
MIT License
|
2 |
+
|
3 |
+
Copyright (c) 2022 Miguel Jaramillo
|
4 |
+
|
5 |
+
Permission is hereby granted, free of charge, to any person obtaining a copy
|
6 |
+
of this software and associated documentation files (the "Software"), to deal
|
7 |
+
in the Software without restriction, including without limitation the rights
|
8 |
+
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
9 |
+
copies of the Software, and to permit persons to whom the Software is
|
10 |
+
furnished to do so, subject to the following conditions:
|
11 |
+
|
12 |
+
The above copyright notice and this permission notice shall be included in all
|
13 |
+
copies or substantial portions of the Software.
|
14 |
+
|
15 |
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
16 |
+
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
17 |
+
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
18 |
+
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
19 |
+
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
20 |
+
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
21 |
+
SOFTWARE.
|
README.md
ADDED
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
title: SpiceIcaroTP
|
3 |
+
emoji: 🌍
|
4 |
+
colorFrom: pink
|
5 |
+
colorTo: gray
|
6 |
+
sdk: gradio
|
7 |
+
sdk_version: 2.9.4
|
8 |
+
app_file: app_deploy.py
|
9 |
+
pinned: false
|
10 |
+
license: mit
|
11 |
+
duplicated_from: mjaramillo/SpiceIcaroTP
|
12 |
+
---
|
13 |
+
|
14 |
+
Check out the configuration reference at https://huggingface.co/docs/hub/spaces#reference
|
TP3.ipynb
ADDED
The diff for this file is too large to render.
See raw diff
|
|
app_deploy.py
ADDED
@@ -0,0 +1,506 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# -*- coding: utf-8 -*-
|
2 |
+
"""tp3__1_-1.ipynb
|
3 |
+
|
4 |
+
Automatically generated by Colaboratory.
|
5 |
+
|
6 |
+
Original file is located at
|
7 |
+
https://colab.research.google.com/drive/1_Sjx5G1BW689ggZJAJ4P7kCZndOobNCp
|
8 |
+
"""
|
9 |
+
|
10 |
+
# Install Gradio
|
11 |
+
#!pip install gradio -q
|
12 |
+
|
13 |
+
# Install timidy
|
14 |
+
#!sudo apt-get install -q -y timidity libsndfile1
|
15 |
+
|
16 |
+
# All the imports to deal with sound data
|
17 |
+
#!pip install pydub numba==0.48 librosa music21
|
18 |
+
|
19 |
+
# Import Libraries
|
20 |
+
|
21 |
+
import gradio as gr
|
22 |
+
import time
|
23 |
+
|
24 |
+
import tensorflow as tf
|
25 |
+
import tensorflow_hub as hub
|
26 |
+
|
27 |
+
import numpy as np
|
28 |
+
import matplotlib.pyplot as plt
|
29 |
+
import librosa
|
30 |
+
from librosa import display as librosadisplay
|
31 |
+
|
32 |
+
import logging
|
33 |
+
import math
|
34 |
+
import statistics
|
35 |
+
import sys
|
36 |
+
|
37 |
+
from IPython.display import Audio, Javascript
|
38 |
+
from scipy.io import wavfile
|
39 |
+
|
40 |
+
from base64 import b64decode
|
41 |
+
|
42 |
+
import music21
|
43 |
+
from pydub import AudioSegment
|
44 |
+
|
45 |
+
logger = logging.getLogger()
|
46 |
+
logger.setLevel(logging.ERROR)
|
47 |
+
|
48 |
+
#print("tensorflow: %s" % tf.__version__)
|
49 |
+
#print("librosa: %s" % librosa.__version__)
|
50 |
+
|
51 |
+
# The audio input file
|
52 |
+
# Now the hardest part: Record your singing! :)
|
53 |
+
|
54 |
+
# We provide four methods to obtain an audio file:
|
55 |
+
|
56 |
+
# 1. Record audio directly in Gradio
|
57 |
+
# 2. Use a file saved on Google Drive
|
58 |
+
|
59 |
+
# Use a file saved on Google Drive
|
60 |
+
#INPUT_SOURCE = 'https://storage.googleapis.com/download.tensorflow.org/data/c-scale-metronome.wav'
|
61 |
+
|
62 |
+
#!wget --no-check-certificate 'https://storage.googleapis.com/download.tensorflow.org/data/c-scale-metronome.wav' -O c-scale.wav
|
63 |
+
|
64 |
+
#uploaded_file_name = 'c-scale.wav'
|
65 |
+
|
66 |
+
#uploaded_file_name
|
67 |
+
|
68 |
+
# Function that converts the user-created audio to the format that the model
|
69 |
+
# expects: bitrate 16kHz and only one channel (mono).
|
70 |
+
|
71 |
+
EXPECTED_SAMPLE_RATE = 16000
|
72 |
+
|
73 |
+
# Funciones #
|
74 |
+
def convert_audio_for_model(user_file, output_file='converted_audio_file.wav'):
|
75 |
+
audio = AudioSegment.from_file(user_file)
|
76 |
+
audio = audio.set_frame_rate(EXPECTED_SAMPLE_RATE).set_channels(1)
|
77 |
+
audio.export(output_file, format="wav")
|
78 |
+
return output_file
|
79 |
+
|
80 |
+
MAX_ABS_INT16 = 32768.0
|
81 |
+
|
82 |
+
def plot_stft(x, sample_rate, show_black_and_white=False):
|
83 |
+
x_stft = np.abs(librosa.stft(x, n_fft=2048))
|
84 |
+
fig, ax = plt.subplots()
|
85 |
+
fig.set_size_inches(20, 10)
|
86 |
+
x_stft_db = librosa.amplitude_to_db(x_stft, ref=np.max)
|
87 |
+
|
88 |
+
if(show_black_and_white):
|
89 |
+
librosadisplay.specshow(data=x_stft_db,
|
90 |
+
y_axis='log',
|
91 |
+
sr=sample_rate,
|
92 |
+
cmap='gray_r')
|
93 |
+
else:
|
94 |
+
librosadisplay.specshow(data=x_stft_db,
|
95 |
+
y_axis='log',
|
96 |
+
sr=sample_rate)
|
97 |
+
|
98 |
+
plt.colorbar(format='%+2.0f dB')
|
99 |
+
|
100 |
+
return fig
|
101 |
+
|
102 |
+
# Loading audio samples from the wav file:
|
103 |
+
#sample_rate, audio_samples = wavfile.read(converted_audio_file, 'rb')
|
104 |
+
|
105 |
+
#fig = plot_stft(audio_samples / MAX_ABS_INT16 , sample_rate=EXPECTED_SAMPLE_RATE)
|
106 |
+
|
107 |
+
# Executing the Model
|
108 |
+
# Loading the SPICE model is easy:
|
109 |
+
model = hub.load("https://tfhub.dev/google/spice/2")
|
110 |
+
|
111 |
+
def plot_pitch_conf(pitch_outputs,confidence_outputs):
|
112 |
+
fig, ax = plt.subplots()
|
113 |
+
fig.set_size_inches(20, 10)
|
114 |
+
plt.plot(pitch_outputs, label='pitch')
|
115 |
+
plt.plot(confidence_outputs, label='confidence')
|
116 |
+
plt.legend(loc="lower right")
|
117 |
+
return fig
|
118 |
+
|
119 |
+
def plot_pitch_conf_notes(confident_pitch_outputs_x,confident_pitch_outputs_y):
|
120 |
+
fig, ax = plt.subplots()
|
121 |
+
fig.set_size_inches(20, 10)
|
122 |
+
ax.set_ylim([0, 1])
|
123 |
+
plt.scatter(confident_pitch_outputs_x, confident_pitch_outputs_y, )
|
124 |
+
plt.scatter(confident_pitch_outputs_x, confident_pitch_outputs_y, c="r")
|
125 |
+
return fig
|
126 |
+
|
127 |
+
def output2hz(pitch_output):
|
128 |
+
# Constants taken from https://tfhub.dev/google/spice/2
|
129 |
+
PT_OFFSET = 25.58
|
130 |
+
PT_SLOPE = 63.07
|
131 |
+
FMIN = 10.0;
|
132 |
+
BINS_PER_OCTAVE = 12.0;
|
133 |
+
cqt_bin = pitch_output * PT_SLOPE + PT_OFFSET;
|
134 |
+
return FMIN * 2.0 ** (1.0 * cqt_bin / BINS_PER_OCTAVE)
|
135 |
+
|
136 |
+
def espectro_notas(audio_samples,EXPECTED_SAMPLE_RATE,confident_pitch_outputs_x,confident_pitch_values_hz):
|
137 |
+
fig, ax = plt.subplots()
|
138 |
+
plot_stft(audio_samples / MAX_ABS_INT16 ,
|
139 |
+
sample_rate=EXPECTED_SAMPLE_RATE, show_black_and_white=True)
|
140 |
+
# Note: conveniently, since the plot is in log scale, the pitch outputs
|
141 |
+
# also get converted to the log scale automatically by matplotlib.
|
142 |
+
plt.scatter(confident_pitch_outputs_x, confident_pitch_values_hz, c="r")
|
143 |
+
return fig
|
144 |
+
|
145 |
+
def hz2offset(freq):
|
146 |
+
# This measures the quantization error for a single note.
|
147 |
+
if freq == 0: # Rests always have zero error.
|
148 |
+
return None
|
149 |
+
# Quantized note.
|
150 |
+
h = round(12 * math.log2(freq / C0))
|
151 |
+
return 12 * math.log2(freq / C0) - h
|
152 |
+
|
153 |
+
def quantize_predictions(group, ideal_offset):
|
154 |
+
# Group values are either 0, or a pitch in Hz.
|
155 |
+
non_zero_values = [v for v in group if v != 0]
|
156 |
+
zero_values_count = len(group) - len(non_zero_values)
|
157 |
+
|
158 |
+
# Create a rest if 80% is silent, otherwise create a note.
|
159 |
+
if zero_values_count > 0.8 * len(group):
|
160 |
+
# Interpret as a rest. Count each dropped note as an error, weighted a bit
|
161 |
+
# worse than a badly sung note (which would 'cost' 0.5).
|
162 |
+
return 0.51 * len(non_zero_values), "Rest"
|
163 |
+
else:
|
164 |
+
# Interpret as note, estimating as mean of non-rest predictions.
|
165 |
+
h = round(
|
166 |
+
statistics.mean([
|
167 |
+
12 * math.log2(freq / C0) - ideal_offset for freq in non_zero_values
|
168 |
+
]))
|
169 |
+
octave = h // 12
|
170 |
+
n = h % 12
|
171 |
+
note = note_names[n] + str(octave)
|
172 |
+
# Quantization error is the total difference from the quantized note.
|
173 |
+
error = sum([
|
174 |
+
abs(12 * math.log2(freq / C0) - ideal_offset - h)
|
175 |
+
for freq in non_zero_values
|
176 |
+
])
|
177 |
+
return error, note
|
178 |
+
|
179 |
+
def get_quantization_and_error(pitch_outputs_and_rests, predictions_per_eighth,
|
180 |
+
prediction_start_offset, ideal_offset):
|
181 |
+
# Apply the start offset - we can just add the offset as rests.
|
182 |
+
pitch_outputs_and_rests = [0] * prediction_start_offset + \
|
183 |
+
pitch_outputs_and_rests
|
184 |
+
# Collect the predictions for each note (or rest).
|
185 |
+
groups = [
|
186 |
+
pitch_outputs_and_rests[i:i + predictions_per_eighth]
|
187 |
+
for i in range(0, len(pitch_outputs_and_rests), predictions_per_eighth)
|
188 |
+
]
|
189 |
+
|
190 |
+
quantization_error = 0
|
191 |
+
|
192 |
+
notes_and_rests = []
|
193 |
+
for group in groups:
|
194 |
+
error, note_or_rest = quantize_predictions(group, ideal_offset)
|
195 |
+
quantization_error += error
|
196 |
+
notes_and_rests.append(note_or_rest)
|
197 |
+
|
198 |
+
return quantization_error, notes_and_rests
|
199 |
+
|
200 |
+
def main(audio):
|
201 |
+
|
202 |
+
# Preparing the audio data
|
203 |
+
# Now we have the audio, let's convert it to the expected format and then
|
204 |
+
# listen to it!
|
205 |
+
# The SPICE model needs as input an audio file at a sampling rate of 16kHz and
|
206 |
+
# with only one channel (mono).
|
207 |
+
# To help you with this part, we created a function(`convert_audio_for_model`)
|
208 |
+
#to convert any wav file you have to the model's expected format:
|
209 |
+
|
210 |
+
|
211 |
+
# Converting to the expected format for the model
|
212 |
+
# in all the input 4 input method before, the uploaded file name is at
|
213 |
+
# the variable uploaded_file_name
|
214 |
+
converted_audio_file = convert_audio_for_model(audio)
|
215 |
+
|
216 |
+
# Loading audio samples from the wav file:
|
217 |
+
sample_rate, audio_samples = wavfile.read(converted_audio_file, 'rb')
|
218 |
+
|
219 |
+
audio_samples = audio_samples / float(MAX_ABS_INT16)
|
220 |
+
|
221 |
+
|
222 |
+
# We now feed the audio to the SPICE tf.hub model to obtain pitch and uncertainty outputs as tensors.
|
223 |
+
model_output = model.signatures["serving_default"](tf.constant(audio_samples, tf.float32))
|
224 |
+
|
225 |
+
pitch_outputs = model_output["pitch"]
|
226 |
+
uncertainty_outputs = model_output["uncertainty"]
|
227 |
+
|
228 |
+
# 'Uncertainty' basically means the inverse of confidence.
|
229 |
+
confidence_outputs = 1.0 - uncertainty_outputs
|
230 |
+
|
231 |
+
|
232 |
+
confidence_outputs = list(confidence_outputs)
|
233 |
+
pitch_outputs = [ float(x) for x in pitch_outputs]
|
234 |
+
|
235 |
+
indices = range(len (pitch_outputs))
|
236 |
+
confident_pitch_outputs = [ (i,p)
|
237 |
+
for i, p, c in zip(indices, pitch_outputs, confidence_outputs) if c >= 0.9 ]
|
238 |
+
confident_pitch_outputs_x, confident_pitch_outputs_y = zip(*confident_pitch_outputs)
|
239 |
+
|
240 |
+
confident_pitch_values_hz = [ output2hz(p) for p in confident_pitch_outputs_y ]
|
241 |
+
|
242 |
+
|
243 |
+
#Plot waves
|
244 |
+
fig1 = plt.figure()
|
245 |
+
plt.plot(audio_samples)
|
246 |
+
|
247 |
+
#Plot
|
248 |
+
fig2 = plot_stft(audio_samples / MAX_ABS_INT16 , sample_rate=EXPECTED_SAMPLE_RATE)
|
249 |
+
|
250 |
+
#Plot Pitch & Confidence
|
251 |
+
fig3 = plot_pitch_conf(pitch_outputs,confidence_outputs)
|
252 |
+
|
253 |
+
|
254 |
+
#Plot Pitch & Confidence Notes
|
255 |
+
fig4 = plot_pitch_conf_notes(confident_pitch_outputs_x,confident_pitch_outputs_y)
|
256 |
+
|
257 |
+
#Plot Espectro + Notes
|
258 |
+
fig5 = espectro_notas(audio_samples,EXPECTED_SAMPLE_RATE,confident_pitch_outputs_x,confident_pitch_values_hz)
|
259 |
+
|
260 |
+
|
261 |
+
# ############################################################################
|
262 |
+
# Converting to musical notes ################################################
|
263 |
+
|
264 |
+
# Now that we have the pitch values, let's convert them to notes!
|
265 |
+
# This is part is challenging by itself. We have to take into account two
|
266 |
+
# things:
|
267 |
+
# 1. the rests (when there's no singing)
|
268 |
+
# 2. the size of each note (offsets)
|
269 |
+
|
270 |
+
# ----------------------------------------------------------------------------
|
271 |
+
### 1: Adding zeros to the output to indicate when there's no singing
|
272 |
+
|
273 |
+
pitch_outputs_and_rests = [
|
274 |
+
output2hz(p) if c >= 0.9 else 0
|
275 |
+
for i, p, c in zip(indices, pitch_outputs, confidence_outputs)
|
276 |
+
]
|
277 |
+
|
278 |
+
# ----------------------------------------------------------------------------
|
279 |
+
### 2: Adding note offsets
|
280 |
+
# When a person sings freely, the melody may have an offset to the absolute
|
281 |
+
# pitch values that notes can represent.
|
282 |
+
# Hence, to convert predictions to notes, one needs to correct for this
|
283 |
+
# possible offset.
|
284 |
+
# This is what the following code computes.
|
285 |
+
|
286 |
+
A4 = 440
|
287 |
+
C0 = A4 * pow(2, -4.75)
|
288 |
+
note_names = ["C", "C#", "D", "D#", "E", "F", "F#", "G", "G#", "A", "A#", "B"]
|
289 |
+
|
290 |
+
def hz2offset(freq):
|
291 |
+
# This measures the quantization error for a single note.
|
292 |
+
if freq == 0: # Rests always have zero error.
|
293 |
+
return None
|
294 |
+
# Quantized note.
|
295 |
+
h = round(12 * math.log2(freq / C0))
|
296 |
+
return 12 * math.log2(freq / C0) - h
|
297 |
+
|
298 |
+
|
299 |
+
# The ideal offset is the mean quantization error for all the notes
|
300 |
+
# (excluding rests):
|
301 |
+
offsets = [hz2offset(p) for p in pitch_outputs_and_rests if p != 0]
|
302 |
+
#print("offsets: ", offsets)
|
303 |
+
off = offsets
|
304 |
+
|
305 |
+
ideal_offset = statistics.mean(offsets)
|
306 |
+
#print("ideal offset: ", ideal_offset)
|
307 |
+
ideal_off = ideal_offset
|
308 |
+
|
309 |
+
# We can now use some heuristics to try and estimate the most likely sequence
|
310 |
+
# of notes that were sung.
|
311 |
+
# The ideal offset computed above is one ingredient - but we also need to know
|
312 |
+
# the speed (how many predictions make, say, an eighth?), and the time offset
|
313 |
+
# to start quantizing. To keep it simple, we'll just try different speeds and
|
314 |
+
# time offsets and measure the quantization error, using in the end the values
|
315 |
+
# that minimize this error.
|
316 |
+
|
317 |
+
def quantize_predictions(group, ideal_offset):
|
318 |
+
# Group values are either 0, or a pitch in Hz.
|
319 |
+
non_zero_values = [v for v in group if v != 0]
|
320 |
+
zero_values_count = len(group) - len(non_zero_values)
|
321 |
+
|
322 |
+
# Create a rest if 80% is silent, otherwise create a note.
|
323 |
+
if zero_values_count > 0.8 * len(group):
|
324 |
+
# Interpret as a rest. Count each dropped note as an error, weighted a bit
|
325 |
+
# worse than a badly sung note (which would 'cost' 0.5).
|
326 |
+
return 0.51 * len(non_zero_values), "Rest"
|
327 |
+
else:
|
328 |
+
# Interpret as note, estimating as mean of non-rest predictions.
|
329 |
+
h = round(
|
330 |
+
statistics.mean([
|
331 |
+
12 * math.log2(freq / C0) - ideal_offset for freq in non_zero_values
|
332 |
+
]))
|
333 |
+
octave = h // 12
|
334 |
+
n = h % 12
|
335 |
+
note = note_names[n] + str(octave)
|
336 |
+
# Quantization error is the total difference from the quantized note.
|
337 |
+
error = sum([
|
338 |
+
abs(12 * math.log2(freq / C0) - ideal_offset - h)
|
339 |
+
for freq in non_zero_values
|
340 |
+
])
|
341 |
+
return error, note
|
342 |
+
|
343 |
+
|
344 |
+
def get_quantization_and_error(pitch_outputs_and_rests, predictions_per_eighth,
|
345 |
+
prediction_start_offset, ideal_offset):
|
346 |
+
# Apply the start offset - we can just add the offset as rests.
|
347 |
+
pitch_outputs_and_rests = [0] * prediction_start_offset + \
|
348 |
+
pitch_outputs_and_rests
|
349 |
+
# Collect the predictions for each note (or rest).
|
350 |
+
groups = [
|
351 |
+
pitch_outputs_and_rests[i:i + predictions_per_eighth]
|
352 |
+
for i in range(0, len(pitch_outputs_and_rests), predictions_per_eighth)
|
353 |
+
]
|
354 |
+
|
355 |
+
quantization_error = 0
|
356 |
+
|
357 |
+
notes_and_rests = []
|
358 |
+
for group in groups:
|
359 |
+
error, note_or_rest = quantize_predictions(group, ideal_offset)
|
360 |
+
quantization_error += error
|
361 |
+
notes_and_rests.append(note_or_rest)
|
362 |
+
|
363 |
+
return quantization_error, notes_and_rests
|
364 |
+
|
365 |
+
|
366 |
+
best_error = float("inf")
|
367 |
+
best_notes_and_rests = None
|
368 |
+
best_predictions_per_note = None
|
369 |
+
|
370 |
+
for predictions_per_note in range(20, 65, 1):
|
371 |
+
for prediction_start_offset in range(predictions_per_note):
|
372 |
+
|
373 |
+
error, notes_and_rests = get_quantization_and_error(
|
374 |
+
pitch_outputs_and_rests, predictions_per_note,
|
375 |
+
prediction_start_offset, ideal_offset)
|
376 |
+
|
377 |
+
if error < best_error:
|
378 |
+
best_error = error
|
379 |
+
best_notes_and_rests = notes_and_rests
|
380 |
+
best_predictions_per_note = predictions_per_note
|
381 |
+
|
382 |
+
# At this point, best_notes_and_rests contains the best quantization.
|
383 |
+
# Since we don't need to have rests at the beginning, let's remove these:
|
384 |
+
#while best_notes_and_rests[0] == 'Rest':
|
385 |
+
# best_notes_and_rests = best_notes_and_rests[1:]
|
386 |
+
# Also remove silence at the end.
|
387 |
+
#while best_notes_and_rests[-1] == 'Rest':
|
388 |
+
# best_notes_and_rests = best_notes_and_rests[:-1]
|
389 |
+
|
390 |
+
# ____________________________________________________________________________
|
391 |
+
# Now let's write the quantized notes as sheet music score!
|
392 |
+
# To do it we will use two libraries: [music21](http://web.mit.edu/music21/) and
|
393 |
+
# [Open Sheet Music Display](https://github.com/opensheetmusicdisplay/opensheetmusicdisplay)
|
394 |
+
# **Note:** for simplicity, we assume here that all notes have the same duration
|
395 |
+
# (a half note).
|
396 |
+
|
397 |
+
# Creating the sheet music score.
|
398 |
+
sc = music21.stream.Score()
|
399 |
+
# Adjust the speed to match the actual singing.
|
400 |
+
bpm = 60 * 60 / best_predictions_per_note
|
401 |
+
#print ('bpm: ', bpm)
|
402 |
+
a = music21.tempo.MetronomeMark(number=bpm)
|
403 |
+
sc.insert(0,a)
|
404 |
+
|
405 |
+
for snote in best_notes_and_rests:
|
406 |
+
d = 'half'
|
407 |
+
if snote == 'Rest':
|
408 |
+
sc.append(music21.note.Rest(type=d))
|
409 |
+
else:
|
410 |
+
sc.append(music21.note.Note(snote, type=d))
|
411 |
+
|
412 |
+
|
413 |
+
# @title [Run this] Helper function to use Open Sheet Music Display (JS code)
|
414 |
+
# to show a music score
|
415 |
+
from IPython.core.display import HTML, Javascript
|
416 |
+
from IPython import display
|
417 |
+
import json, random
|
418 |
+
|
419 |
+
def showScore(score):
|
420 |
+
xml = open(score.write('musicxml')).read()
|
421 |
+
showMusicXML(xml)
|
422 |
+
|
423 |
+
def showMusicXML(xml):
|
424 |
+
DIV_ID = "OSMD_div"
|
425 |
+
a = display(HTML('<div id="'+DIV_ID+'">loading OpenSheetMusicDisplay</div>'))
|
426 |
+
script = """
|
427 |
+
var div_id = {{DIV_ID}};
|
428 |
+
function loadOSMD() {
|
429 |
+
return new Promise(function(resolve, reject){
|
430 |
+
if (window.opensheetmusicdisplay) {
|
431 |
+
return resolve(window.opensheetmusicdisplay)
|
432 |
+
}
|
433 |
+
// OSMD script has a 'define' call which conflicts with requirejs
|
434 |
+
var _define = window.define // save the define object
|
435 |
+
window.define = undefined // now the loaded script will ignore requirejs
|
436 |
+
var s = document.createElement( 'script' );
|
437 |
+
s.setAttribute( 'src', "https://cdn.jsdelivr.net/npm/opensheetmusicdisplay@0.7.6/build/opensheetmusicdisplay.min.js" );
|
438 |
+
//s.setAttribute( 'src', "/custom/opensheetmusicdisplay.js" );
|
439 |
+
s.onload=function(){
|
440 |
+
window.define = _define
|
441 |
+
resolve(opensheetmusicdisplay);
|
442 |
+
};
|
443 |
+
document.body.appendChild( s ); // browser will try to load the new script tag
|
444 |
+
})
|
445 |
+
}
|
446 |
+
loadOSMD().then((OSMD)=>{
|
447 |
+
window.openSheetMusicDisplay = new OSMD.OpenSheetMusicDisplay(div_id, {
|
448 |
+
drawingParameters: "compacttight"
|
449 |
+
});
|
450 |
+
openSheetMusicDisplay
|
451 |
+
.load({{data}})
|
452 |
+
.then(
|
453 |
+
function() {
|
454 |
+
openSheetMusicDisplay.render();
|
455 |
+
}
|
456 |
+
);
|
457 |
+
})
|
458 |
+
""".replace('{{DIV_ID}}',DIV_ID).replace('{{data}}',json.dumps(xml))
|
459 |
+
display(Javascript(script))
|
460 |
+
return a
|
461 |
+
|
462 |
+
# rendering the music score
|
463 |
+
###partitura = showScore(sc)
|
464 |
+
#print(best_notes_and_rests)
|
465 |
+
|
466 |
+
|
467 |
+
|
468 |
+
# ____________________________________________________________________________
|
469 |
+
# Let's convert the music notes to a MIDI file and listen to it.
|
470 |
+
# To create this file, we can use the stream we created before.
|
471 |
+
|
472 |
+
# Saving the recognized musical notes as a MIDI file
|
473 |
+
##converted_audio_file_as_midi = converted_audio_file[:-4] + '.mid'
|
474 |
+
##fp = sc.write('midi', fp=converted_audio_file_as_midi)
|
475 |
+
|
476 |
+
##wav_from_created_midi = converted_audio_file_as_midi.replace(' ', '_') + "_midioutput.wav"
|
477 |
+
#print(wav_from_created_midi)
|
478 |
+
|
479 |
+
# To listen to it on colab, we need to convert it back to wav. An easy way of
|
480 |
+
# doing that is using Timidity.
|
481 |
+
|
482 |
+
#!timidity $converted_audio_file_as_midi -Ow -o $wav_from_created_midi
|
483 |
+
return converted_audio_file, fig1, fig2, fig3, fig4,fig5, bpm, best_notes_and_rests#, wav_from_created_midi
|
484 |
+
#return converted_audio_file, fig1, fig2, fig3, fig4,fig5, bpm, best_notes_and_rests, partitura, wav_from_created_midi
|
485 |
+
|
486 |
+
link = "https://www.tensorflow.org/hub/tutorials/spice?hl=es-419&authuser=2"
|
487 |
+
|
488 |
+
iface = gr.Interface(
|
489 |
+
fn=main,
|
490 |
+
title= "Trabajo Práctico N°3 - Detección de tono con SPICE",
|
491 |
+
description="Implementación de Modelo con GitHub + Hugging Face🤗-- 🔊✅ " + "Basado en: " + link,
|
492 |
+
inputs = [gr.inputs.Audio(source= "microphone" , type="filepath",label="Ingrese Audio")],
|
493 |
+
outputs= [gr.outputs.Audio(label="Audio Original"),
|
494 |
+
gr.outputs.Plot(type="auto",label="Gráfico de Frecuencias"),
|
495 |
+
gr.outputs.Plot(type="auto",label="Especto"),
|
496 |
+
gr.outputs.Plot(type="auto",label="Pitch Confidence"),
|
497 |
+
gr.outputs.Plot(type="auto",label="Notas"),
|
498 |
+
gr.outputs.Plot(type="auto",label="Espectro+Notas"),
|
499 |
+
gr.outputs.Textbox(label="bpm"),
|
500 |
+
gr.outputs.Textbox(label="partitura")],#,
|
501 |
+
#gr.outputs.Textbox(type="html",label="partitura1"),
|
502 |
+
#gr.outputs.Audio(label="midi")],
|
503 |
+
interpretation = "default",
|
504 |
+
)
|
505 |
+
|
506 |
+
iface.launch(debug=True)
|
requirements.txt
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
numpy==1.21.2
|
2 |
+
tensorflow==2.8.0
|
3 |
+
tensorflow_hub==0.12.0
|
4 |
+
matplotlib==3.5.1
|
5 |
+
statistics==1.0.3.5
|
6 |
+
ipython==8.3.0
|
7 |
+
scipy==1.8.0
|
8 |
+
music21==7.3.3
|
9 |
+
pydub
|
10 |
+
librosa==0.9.1
|