FileNotFoundError: [Errno 2] No such file or directory: 'ffmpeg'

[JeanVel]

Hello, I am using a Mac with a M1 chip and I have an issue.
I tried to install whisper to transcribe an audio, I followed the instruction, I managed to install the package whisper, and ffmpeg (using homebrew). But when I use the following command : transcription = model.transcribe(audio_path, fp16=False), I get the following error : FileNotFoundError: [Errno 2] No such file or directory: 'ffmpeg'. Does anyone has an idea of how I could fix this ? Thank you

[yasersakkaf]

Facing the same issue, any solutions yet?

[SAYUK09]

Encountered this on ubuntu system and the following solved it for me.

Step 1:
sudo apt update && sudo apt install ffmpeg

Step 2:
sudo apt install ffmpeg

Hope this helps.

[Rentetier]

I have the same issue, any solutions?

[SAYUK09]

This helped me to solve it on ubuntu
Just encountered the same error on ubuntu, and solved it using the following way.

Step 1:
sudo apt update && sudo apt install ffmpeg

Step 2:
sudo apt install ffmpeg

[glangford]

ffmpeg must be in your PATH for whisper to use it. In terminal try which ffmpeg to verify. Otherwise, see here for installation options:

• https://superuser.com/questions/624561/install-ffmpeg-on-os-x

Or this prior discussion may be helpful:

• M1 Mac Installation Troubleshooting Guide for Whisper, Ffmpeg and Brew #1238

[guus-vrv]

This worked for me, thanks!

[MissUAA]

This worked for me, too. Thank you !

[minerb007]

once I installed FFMPEG and add it to the path, I still had issues running it via terminal in VS Code, but running it through windows command prompt solved my issues

[Codie-Petersen]

This user's problem is a bit different than yours. They are doing it for Mac. Sometimes you have to reload VSCode to get the path to load correctly. Definitely make sure you reload the specific terminal you were using in VSCode because it won't know anything about the updated Path unless it was started after Path was updated.

For any other windows users here are the install instructions for FFMPEG:
https://phoenixnap.com/kb/ffmpeg-windows

[paranoid2droid]

In my case, ffmpeg was installed in the base environment. When I switched to another conda environment, the same error appears. If you are also using conda, the simplest solution might be just to install ffmpeg via conda:
conda install -c conda-forge ffmpeg

[SAYUK09]

Just encountered the same error on ubuntu, and solved it using the following way.

• Step 1:
  sudo apt update && sudo apt install ffmpeg

• Step 2:
  sudo apt install ffmpeg

Hope this helps.

[Yuanyuan-888]

Hi, everyone,

I am quite confusing with this error.

As long as I am using a40 gpu in my university HPC to test with whisper-large model, it shows this ffmpeg error! But when I used v100 gpu, this error is gone, and the whisper-large can successfully decode something.

When I use "which ffmpeg", it shows the location "/usr/bin/ffmpeg"
and in this conda envirnment, when I "import ffmpeg", it is also successful.

By the way, on our hpc system, we do not have sudo accessment TT

If anyone has some idea about why this error is gpu dependent, please do leave me a message, need you help!

[okineadev]

Can you provide logs?

[Yuanyuan-888]

2024-04-27T08:27:44 (evaluate_asr.sh:106:main) scripts/utils/evaluate_asr.sh --whisper_tag large --nj 2 --gpu_inference true --stage 2 --stop_stage 3 --cleaner whisper_basic --hyp_cleaner whisper_basic --decode_options {language: nl, task: transcribe, temperature: 0, beam_size: 10, fp16: False} --gt_text dump/raw/dev_s/text dump/raw/dev_s/wav.scp whisper-large_outputs_nl_0424/dev_s
2024-04-27T08:27:50 (evaluate_asr.sh:160:main) stage 2: ASR inference
2024-04-27T08:27:50 (evaluate_asr.sh:191:main) Decoding started... log: 'whisper-large_outputs_nl_0424/dev_s/logdir/asr_inference.*.log'
run.pl: job failed, log is in whisper-large_outputs_nl_0424/dev_s/logdir/asr_inference.1.log

python3 pyscripts/utils/evaluate_whisper_inference.py --ngpu 1 --data_path_and_name_and_type dump/raw/dev_s/wav.scp --key_file whisper-large_outputs_nl_0424/dev_s/logdir/keys.1.scp --model_tag large --model_dir whisper-large_outputs_nl_0424/dev_s/models --output_dir whisper-large_outputs_nl_0424/dev_s/logdir/output.1 --decode_options "{language: nl, task: transcribe, temperature: 0, beam_size: 10, fp16: False}"

Started at Sat Apr 27 08:27:50 CEST 2024

/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/site-packages/whisper/timing.py:57: NumbaDeprecationWarning: ^[[1mThe 'nopython' keyword argument was not supplied to the 'numba.jit' decorator. The implicit default value for this argument is currently False, but it will be changed to True in Numba 0.59.0. See https://numba.readthedocs.io/en/stable/reference/deprecation.html#deprecation-of-object-mode-fall-back-behaviour-when-using-jit for details.^[[0m
def backtrace(trace: np.ndarray):
/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/bin/python3 pyscripts/utils/evaluate_whisper_inference.py --ngpu 1 --data_path_and_name_and_type dump/raw/dev_s/wav.scp --key_file whisper-large_outputs_nl_0424/dev_s/logdir/keys.1.scp --model_tag large --model_dir whisper-large_outputs_nl_0424/dev_s/models --output_dir whisper-large_outputs_nl_0424/dev_s/logdir/output.1 --decode_options '{language: nl, task: transcribe, temperature: 0, beam_size: 10, fp16: False}'
Traceback (most recent call last):
File "pyscripts/utils/evaluate_whisper_inference.py", line 195, in
main()
File "pyscripts/utils/evaluate_whisper_inference.py", line 191, in main
inference(**kwargs)
File "pyscripts/utils/evaluate_whisper_inference.py", line 103, in inference
results = speech2text(os.path.abspath(audio_file.strip()), **decode_options)
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
return func(*args, **kwargs)
File "pyscripts/utils/evaluate_whisper_inference.py", line 50, in call
result = self.model.transcribe(speech, **decode_options)
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/site-packages/whisper/transcribe.py", line 121, in transcribe
mel = log_mel_spectrogram(audio, padding=N_SAMPLES)
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/site-packages/whisper/audio.py", line 130, in log_mel_spectrogram
audio = load_audio(audio)
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/site-packages/whisper/audio.py", line 46, in load_audio
ffmpeg.input(file, threads=0)
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/site-packages/ffmpeg/_run.py", line 313, in run
process = run_async(
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/site-packages/ffmpeg/_run.py", line 284, in run_async
return subprocess.Popen(
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/subprocess.py", line 858, in init
self._execute_child(args, executable, preexec_fn, close_fds,
File "/tudelft.net/staff-bulk/ewi/insy/SpeechLab/yuanyuanzhang/espnet_yyy/espnet/tools/miniconda/envs/espnet/lib/python3.8/subprocess.py", line 1720, in _execute_child
raise child_exception_type(errno_num, err_msg, err_filename)
FileNotFoundError: [Errno 2] No such file or directory: 'ffmpeg'

Accounting: time=107 threads=1

Ended (code 1) at Sat Apr 27 08:29:37 CEST 2024, elapsed time 107 seconds
srun: error: gpu16: task 0: Exited with exit code 1
srun: launch/slurm: _step_signal: Terminating StepId=9923822.0

This is the error infor on a a40 gpu. but when I decode with the same code on v100, then no error

[Yuanyuan-888]

Hi, I figured out, I think it fails because on my gpu nodes, it did not have ffmpeg command line interface.

[Yuanyuan-888]

Hi, I still have a question, when directly use whisper-large model to decode, in the code miniconda/envs/espnet/lib/python3.8/site-packages/whisper/audio.py

it uses ffmpeg to load the audio but not torchaudio to load the audio.

def load_audio(file: str, sr: int = SAMPLE_RATE):
"""
Open an audio file and read as mono waveform, resampling as necessary

Parameters
----------
file: str
    The audio file to open

sr: int
    The sample rate to resample the audio if necessary

Returns
-------
A NumPy array containing the audio waveform, in float32 dtype.
"""
try:
    # This launches a subprocess to decode audio while down-mixing and resampling as necessary.
    # Requires the ffmpeg CLI and `ffmpeg-python` package to be installed.
    out, _ = (
        ffmpeg.input(file, threads=0)
        .output("-", format="s16le", acodec="pcm_s16le", ac=1, ar=sr)
        .run(cmd=["ffmpeg", "-nostdin"], capture_stdout=True, capture_stderr=True)
    )
except ffmpeg.Error as e:
    raise RuntimeError(f"Failed to load audio: {e.stderr.decode()}") from e

return np.frombuffer(out, np.int16).flatten().astype(np.float32) / 32768.0

If I use torchaudio to replace ffmpeg,

import torchaudio
import torch

def load_audio(file: str, sr: int = SAMPLE_RATE):
"""
Open an audio file and read as mono waveform, resampling as necessary

Parameters
----------
file: str
    The audio file to open

sr: int
    The sample rate to resample the audio if necessary

Returns
-------
A NumPy array containing the audio waveform, in float32 dtype.
"""
try:
    # Load audio using torchaudio
    waveform, _ = torchaudio.load(file, normalize=True, channels_first=True, num_frames=-1)
    
    # Resample if necessary
    if sr != SAMPLE_RATE:
        waveform = torchaudio.transforms.Resample(orig_freq=SAMPLE_RATE, new_freq=sr)(waveform)
    
except Exception as e:
    raise RuntimeError(f"Failed to load audio: {e}") from e

# Convert to NumPy array and return
return waveform.numpy().astype(np.float32)

I think these are the same.