added output path for parallel inference

examples/asr/transcribe_speech_parallel.py

@@ -185,7 +185,7 @@ def main(cfg: ParallelTranscriptionConfig):
     pred_text_list = []
     text_list = []
     if is_global_rank_zero():
-        output_file = os.path.join(cfg.output_path, f"predictions_all.json")
+        output_file = os.path.join(cfg.output_path, cfg.output_filename)
         logging.info(f"Prediction files are being aggregated in {output_file}.")
         with open(output_file, 'w') as outf:
             for rank in range(trainer.world_size):
@@ -198,6 +198,7 @@ def main(cfg: ParallelTranscriptionConfig):
                         text_list.append(item["text"])
                         outf.write(json.dumps(item) + "\n")
                         samples_num += 1
+                os.remove(input_file)
         wer_cer = word_error_rate(hypotheses=pred_text_list, references=text_list, use_cer=cfg.use_cer)
         logging.info(
             f"Prediction is done for {samples_num} samples in total on all workers and results are aggregated in {output_file}."