Trying out DeepSpeech on a Raspberry Pi 4

January 3rd, 2020

Deep Speech is an open speech-to-text engine by Mozilla. Speech synthesis and Speech to text are fun to try out, and I read that it could run on a Raspberry Pi4 with ease on one core, so I decided to give it a try.

The Raspberry Pi version is using Google’s TensorFlow Lite for an implementation of Baidu’s DeepSpeech architecture.

Installing it on a Raspberry 4 Buster distribution was not straightforward. First I read instructions on the Github page and tried to download and install the git version and, but I ran into problems. It was taking ages and I ran into the famous `wheels` problem.

Failed building wheel for scipy

After tweaking and trying a few times, i gave up on the Github version and tried the instructions here, but also that was a bumpy road. But success waits in the end.

Let’s go, how to install DeepSpeech on the RPI4

Create a dev directory:

mkdir dev
cd dev

Create a Python Virtual environment.

python3 -m venv deepspeech-train-venv

Activate the virtual environment

source dev/deepspeech-train-venv/bin/activate

Create the deepspeech directory

mkdir deepspeech
cd deepspeech

Install deepspeech

pip install deepspeech

Download pre-trained English model

curl -LO
tar xvf deepspeech-0.6.0-models.tar.gz

Download example audio files

curl -LO
tar xvf audio-0.6.0.tar.gz

Done, run, well , eh, I tried to run the example on the instruction page

deepspeech --model deepspeech-0.6.0-models/output_graph.pbmm --lm deepspeech-0.6.0-models/lm.binary --trie deepspeech-0.6.0-models/trie --audio audio/2830-3980-0043.wav

Errors!?!  I installed a missing dependency:

sudo apt install libatlas3-base

Still errors

ModuleNotFoundError: No module named 'numpy.core._multiarray_umath'

So I check if I had numpy installed

pip install numpy
Looking in indexes:,
Requirement already satisfied: numpy in /home/pi/dev/deepspeech-train-venv/lib/python3.7/site-packages (1.15.4)

I decided to update numpy:

pip install --upgrade numpy
Looking in indexes:,
Collecting numpy
Using cached
tensorboard 2.0.2 has requirement setuptools>=41.0.0, but you'll have setuptools 40.8.0 which is incompatible.
Installing collected packages: numpy
Found existing installation: numpy 1.15.4
Uninstalling numpy-1.15.4:
Successfully uninstalled numpy-1.15.4
Successfully installed numpy-1.18.0

So i decided to update setuptools too:

pip install --upgrade setuptools
Looking in indexes:,
Collecting setuptools
Using cached
Installing collected packages: setuptools
Found existing installation: setuptools 40.8.0
Uninstalling setuptools-40.8.0:
Successfully uninstalled setuptools-40.8.0
Successfully installed setuptools-44.0.0

I tried to run the example on the instruction page again
# Transcribe an audio file

deepspeech --model deepspeech-0.6.0-models/output_graph.pbmm --lm deepspeech-0.6.0-models/lm.binary --trie deepspeech-0.6.0-models/trie --audio audio/2830-3980-0043.wav

Another error

Loading model from file deepspeech-0.6.0-models/output_graph.pbmm
TensorFlow: v1.14.0-21-ge77504a
DeepSpeech: v0.6.0-0-g6d43e21
ERROR: Model provided has model identifier '='+;', should be 'TFL3'

Didn’t work. I needed to change the model to `tflite`

deepspeech --model deepspeech-0.6.0-models/output_graph.tflite --lm deepspeech-0.6.0-models/lm.binary --trie deepspeech-0.6.0-models/trie --audio audio/2830-3980-0043.wav

Success in the end!

Loading model from file deepspeech-0.6.0-models/output_graph.tflite
TensorFlow: v1.14.0-21-ge77504a
DeepSpeech: v0.6.0-0-g6d43e21
INFO: Initialized TensorFlow Lite runtime.
Loaded model in 0.0019s.
Running inference.
why should one hault on the way
Inference took 4.091s for 2.735s audio file.

Then I played the audio-file:

aplay audio/4507-16021-0012.wav

Must say DeepSpeech is much smarter then me, I couldn’t understand it:
why should one hault on the way

BTW good question. No I need another engine to answer that!

Way to go, folks.


3 Responses to “Trying out DeepSpeech on a Raspberry Pi 4”

  1. Upgrade deepspeech 0.7 on a Raspberry Pi 4 - Says:

    […] an earlier post I described how to install deepspeech on a Raspberry Pi 4. That wasn’t exactly a really smooth install, but I managed in the […]

  2. Max Says:


    thank you for describing your installation process.
    Are you actually using the setup now and how well does it work on the RPi4 and how well is the speech recognition currently?

    Thank You!

  3. webonomic Says:

    To be honest, I haven’t had much time to play around with it, because I lent my RPI3 to a friend (which I need for recording audio)

    The examples are working fine though 😉

Leave a Reply