Keep the custom fields in libriheavy manifest (#1719)

egs/libriheavy/ASR/local/prepare_manifest.py

@@ -29,17 +29,21 @@ def simple_cleanup(text: str) -> str:
 
 # Assign text of the supervisions and remove unnecessary entries.
 def main():
-    assert len(sys.argv) == 3, "Usage: ./local/prepare_manifest.py INPUT OUTPUT_DIR"
+    assert (
+        len(sys.argv) == 4
+    ), "Usage: ./local/prepare_manifest.py INPUT OUTPUT_DIR KEEP_CUSTOM_FIELDS"
     fname = Path(sys.argv[1]).name
     oname = Path(sys.argv[2]) / fname
+    keep_custom_fields = bool(sys.argv[3])
     with gzip.open(sys.argv[1], "r") as fin, gzip.open(oname, "w") as fout:
         for line in fin:
             cut = json.loads(line)
             cut["supervisions"][0]["text"] = simple_cleanup(
                 cut["supervisions"][0]["custom"]["texts"][0]
             )
-            del cut["supervisions"][0]["custom"]
-            del cut["custom"]
+            if not keep_custom_fields:
+                del cut["supervisions"][0]["custom"]
+                del cut["custom"]
             fout.write((json.dumps(cut) + "\n").encode())
 
 

egs/libriheavy/ASR/prepare.sh

@@ -29,6 +29,11 @@ export CUDA_VISIBLE_DEVICES=""
 #     - speech
 dl_dir=$PWD/download
 
+# If you want to do PromptASR experiments, please set it to True
+# as this will keep the texts and pre_text information required for
+# the training of PromptASR.
+keep_custom_fields=False
+
 . shared/parse_options.sh || exit 1
 
 # vocab size for sentence piece models.
@@ -134,7 +139,7 @@ if [ $stage -le 3 ] && [ $stop_stage -ge 3 ]; then
   for subset in small medium large dev test_clean test_other; do
     if [ ! -e $manifests_dir/libriheavy_cuts_${subset}.jsonl.gz ]; then
       log "Prepare manifest for subset : ${subset}"
-      ./local/prepare_manifest.py $dl_dir/libriheavy/libriheavy_cuts_${subset}.jsonl.gz $manifests_dir
+      ./local/prepare_manifest.py $dl_dir/libriheavy/libriheavy_cuts_${subset}.jsonl.gz $manifests_dir $keep_custom_fields
     fi
   done
 fi