From 521842deea720713176bcafabc8674d40b6f5878 Mon Sep 17 00:00:00 2001
From: Anas Abou Allaban <aabouallaban@pm.me>
Date: Thu, 15 Oct 2020 11:22:39 -0400
Subject: [PATCH] Convert channels for CV2 dataset

When running a training session on the CV2 dataset, it is possible to get the following error:

```
ValueError: Mono-channel audio required
```

This makes the [pysox Transformer](https://pysox.readthedocs.io/en/latest/api.html#sox.transform.Transformer.convert) also convert the channels.
---
 bin/import_cv2.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/bin/import_cv2.py b/bin/import_cv2.py
index d6c8c192..19a5741c 100755
--- a/bin/import_cv2.py
+++ b/bin/import_cv2.py
@@ -27,6 +27,7 @@ from ds_ctcdecoder import Alphabet
 
 FIELDNAMES = ["wav_filename", "wav_filesize", "transcript"]
 SAMPLE_RATE = 16000
+CHANNELS = 1
 MAX_SECS = 10
 PARAMS = None
 FILTER_OBJ = None
@@ -179,7 +180,7 @@ def _preprocess_data(tsv_dir, audio_dir, space_after_every_character=False):
 def _maybe_convert_wav(mp3_filename, wav_filename):
     if not os.path.exists(wav_filename):
         transformer = sox.Transformer()
-        transformer.convert(samplerate=SAMPLE_RATE)
+        transformer.convert(samplerate=SAMPLE_RATE, n_channels=CHANNELS)
         try:
             transformer.build(mp3_filename, wav_filename)
         except sox.core.SoxError: