sox speed pertrub shape (T, C), float64, process using int32

deepspeech/frontend/audio.py

@@ -351,7 +351,9 @@ class AudioSegment(object):
         tfm.set_globals(multithread=False)
         tfm.speed(speed_rate)
         self._samples = tfm.build_array(
-            input_array=self._samples, sample_rate_in=self._sample_rate).copy()
+            input_array=self._samples,
+            sample_rate_in=self._sample_rate).squeeze(-1).astype(
+                np.float32).copy()
 
     def normalize(self, target_db=-20, max_gain_db=300.0):
         """Normalize audio to be of the desired RMS value in decibels.

examples/librispeech/s0/conf/augmentation.json

@@ -1,4 +1,13 @@
 [
+  {
+    "type": "speed",
+    "params": {
+      "min_speed_rate": 0.9,
+      "max_speed_rate": 1.1,
+      "num_rates": 3
+    },
+    "prob": 0.0
+  },
   {
     "type": "shift",
     "params": {

examples/tiny/s1/conf/augmentation.json

@@ -6,7 +6,7 @@
       "max_speed_rate": 1.1,
       "num_rates": 3
     },
-    "prob": 0.0
+    "prob": 1.0
   },
   {
     "type": "shift",