fix(asr): preserve Unicode in transcription JSON

Signed-off-by: Sylvester Kaczmarek <assistant@SylvesterKaczmarek.com>
Signed-off-by: Sylvester Kaczmarek <16242628+sylvesterkaczmarek@users.noreply.github.com>

Author: sylvesterkaczmarek

nemo/collections/asr/parts/utils/transcribe_utils.py

@@ -536,7 +536,7 @@ def write_transcription(
                         item['pred_lang_chars'] = transcription.langs_chars
                     if not cfg.decoding.beam.return_best_hypothesis:
                         item['beams'] = beams[idx]
-                f.write(json.dumps(item) + "\n")
+                f.write(json.dumps(item, ensure_ascii=False) + "\n")
         else:
             with open(cfg.dataset_manifest, 'r', encoding='utf-8') as fr:
                 for idx, line in enumerate(fr):
@@ -573,7 +573,7 @@ def write_transcription(
 
                         if not cfg.decoding.beam.return_best_hypothesis:
                             item['beams'] = beams[idx]
-                    f.write(json.dumps(item) + "\n")
+                    f.write(json.dumps(item, ensure_ascii=False) + "\n")
 
     return cfg.output_filename, pred_text_attr_name