Bug: blank lines in wav.scp / jsonl file lists crash prepare_data_iterator (IndexError / JSONDecodeError) #3757

🐛 Bug

AutoModel.generate() with a file list (wav.scp, text.txt, file.jsonl, ...) crashes if the list contains a blank or whitespace-only line. prepare_data_iterator() in funasr/auto/auto_model.py iterates over every line without skipping empty ones:

• the id \t data branch does lines = line.strip().split(maxsplit=1) and then indexes lines[0], so an empty line raises IndexError: list index out of range;
• the .jsonl branch calls json.loads(line.strip()) on the empty line and raises json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0).

Blank lines are easy to produce in practice — lists concatenated from several sources, lists edited by hand, or files written with an extra separator line — and one of them aborts the whole batch instead of being ignored.

To Reproduce

1. Install from source: pip install -e .
2. Run the snippet below (it calls prepare_data_iterator directly, so no model weights or download are needed).
3. See IndexError for the .scp list and JSONDecodeError for the .jsonl list.

Code sample

from funasr.auto.auto_model import prepare_data_iterator

open("wav.scp", "w", encoding="utf-8").write("utt1 /a.wav\n\nutt2 /b.wav\n")
open("data.jsonl", "w", encoding="utf-8").write(
    '{"key": "k1", "source": "/a.wav"}\n\n{"key": "k2", "source": "/b.wav"}\n'
)

prepare_data_iterator("wav.scp")     # IndexError: list index out of range
prepare_data_iterator("data.jsonl")  # json.decoder.JSONDecodeError

The same crash reaches users through model.generate(input="wav.scp").

Expected behavior

Blank lines carry no utterance, so they should be skipped; the two real entries should come back as (['utt1', 'utt2'], ['/a.wav', '/b.wav']) and (['k1', 'k2'], ['/a.wav', '/b.wav']), exactly as they do when the blank line is deleted manually.

Error logs

Traceback (most recent call last):
  File "repro.py", line 8, in <module>
    prepare_data_iterator("wav.scp")
  File ".../funasr/auto/auto_model.py", line 479, in prepare_data_iterator
    data = lines[1] if len(lines) > 1 else lines[0]
IndexError: list index out of range

Traceback (most recent call last):
  File "repro.py", line 9, in <module>
    prepare_data_iterator("data.jsonl")
  File ".../funasr/auto/auto_model.py", line 474, in prepare_data_iterator
    lines = json.loads(line.strip())
json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)

Environment

• OS: Windows 11 26200 (the file-list reader is pure Python, so this reproduces on Linux too)
• Python version: 3.11.14
• FunASR version: source main at 66d7a4c264a5993a2a63ed00c1f402c296ee521a
• ModelScope version: not used by this code path
• PyTorch / torchaudio version: 2.14.1+cpu
• Install method (pip, source, Docker): source, pip install -e . equivalent
• Device (cuda, cpu, mps): cpu
• GPU model: n/a
• CUDA/cuDNN version: n/a
• Docker image tag, if used: n/a

Audio details

No audio is needed — the crash happens while parsing the list file itself.