View a spectrogram of the audio

Author: lucasw

audio_to_float/launch/spectrogram.launch

@@ -0,0 +1,22 @@
+<?xml version="1.0"?>
+<launch>
+
+  <arg name="float_sample_rate" default="16000"/>
+  <node name="gen_float" pkg="float_to_audio" type="gen_float.py"
+      output="screen" >
+    <param name="sample_rate" value="$(arg float_sample_rate)"/>
+  </node>
+
+  <node name="spectrogram" pkg="audio_to_float" type="spectrogram.py"
+      output="screen" >
+    <param name="sample_rate" value="$(arg float_sample_rate)"/>
+  </node>
+
+  <node name="view_input" pkg="audio_to_float" type="view.py">
+    <remap from="decoded" to="samples"/>
+    <remap from="image" to="image"/>
+    <param name="fade1" value="0.5"/>
+    <param name="fade2" value="0.5"/>
+  </node>
+
+</launch>

audio_to_float/scripts/spectrogram.py

@@ -0,0 +1,50 @@
+#!/usr/bin/env python
+
+import collections
+import cv2
+import numpy as np
+import rospy
+
+# from audio_common_msgs.msg import AudioData
+from cv_bridge import CvBridge, CvBridgeError
+from scipy import signal
+from sensor_msgs.msg import ChannelFloat32, Image
+
+
+class View():
+    def __init__(self):
+        self.bridge = CvBridge()
+        self.buffer_len = rospy.get_param("~buffer_len", 2**16)
+        self.buffer = collections.deque(maxlen=self.buffer_len)
+        self.sample_rate = rospy.get_param("~sample_rate", 44100)
+        # self.window = 256
+        self.im = None
+        self.pub = rospy.Publisher("image_spectrogram", Image, queue_size=1)
+        self.sub = rospy.Subscriber("samples", ChannelFloat32,
+                                    self.audio_callback, queue_size=1)
+        self.timer = rospy.Timer(rospy.Duration(0.2), self.update)
+
+    def audio_callback(self, msg):
+        for i in range(len(msg.values)):
+            self.buffer.append(msg.values[i])
+
+    def update(self, event):
+        if len(self.buffer) < self.buffer_len:
+            return
+        samples = np.asarray(self.buffer)
+        f, t, Sxx = signal.spectrogram(samples, self.sample_rate)
+        # TODO(lucasw) is there a standard spectrogram conversion?
+        Sxx = np.log(1.0 + Sxx * 2**16)
+        mins = np.min(Sxx)
+        maxs = np.max(Sxx)
+        Sxx -= mins
+        print Sxx.shape, mins, maxs
+        self.im = (Sxx * 50).astype(np.uint8)
+        # self.im[y0:y1+1, i, :] = 255
+        self.pub.publish(self.bridge.cv2_to_imgmsg(self.im, "mono8"))
+        # rospy.signal_shutdown("")
+
+if __name__ == '__main__':
+    rospy.init_node('spectrogram')
+    view = View()
+    rospy.spin()