Made in Vancouver, Canada by Picovoice

This is a minimalist and extensible framework for benchmarking different speech-to-text engines. It has been developed and tested on Ubuntu with Python3.

• Background
• Data
• Metrics
  • Word Error Rate
  • Real Time Factor
  • Memory
• Speech-to-Text Engines
  • Mozilla DeepSpeech
  • Picovoice Cheetah
  • PocketSphinx
• Usage
  • Word Error Rate Measurement
  • Real Time Factor Measurement
  • Memory Usage Measurement
• Results
• License

This framework has been developed by Picovoice as part of the project Cheetah. Cheetah is Picovoice's speech-to-text engine specifically designed for IoT applications. Deep learning has been the main driver in recent improvements in speech recognition. But due to stringent compute/storage limitations of IoT platforms it is most beneficial to the cloud-based engines. Picovoice's proprietary deep learning technology enables transferring these improvements to IoT platforms with much lower CPU/memory footprint. The goal is to be able to run Cheetah on any platform with a C Compiler and a few MB of memory.

This framework enabled us to measure our progress in improving Cheetah and also compare its performance with existing solutions.

Mozilla Common Voice and LibriSpeech datasets are used for benchmarking. Only the cv-valid-test portion of Common Voice dataset is used to allow engines to use the train portion of the dataset. Since the Common Voice dataset is community-verified we only use examples that have no downvotes and at least two upvotes. Similarly, We use the test-clean portion of LibriSpeech dataset to allow engines use the train portions.

Three different metrics are measured.

Word error rate is defined as the Levenstein distance between words in reference transcript and words in the output of the speech-to-text engine to the number of words in reference transcript.

Real-time factor (RTF) is measured as the ratio of CPU (processing) time in seconds to the length of the input speech file in seconds. A speech-to-text engine with lower RTF is computationally more efficient

The amount of heap memory used.

All engines below run fully on-device (no cloud connection needed).

Mozilla DeepSpeech is an open-source implementation of Baidu's DeepSpeech by Mozilla.

Cheetah is a speech-to-text engine developed using Picovoice's proprietary deep learning technology. It works offline and is supported on a growing number of mobile/embedded platforms including Android, iOS, and Raspberry Pi.

PocketSphinx works offline and can run on embedded platforms such as Raspberry Pi.

Below is information on how to use this framework to benchmark engines mentioned above. First, make sure that you have already installed DeepSpeech and PocketSphinx on your machine following instructions on their official pages. Then download Common Voice dataset and test-clean portion of LibriSpeech.

WER can be measured by running the following command from the root of the repository. DATASET_TYPE is either commonvoice or librispeech. DATASET_PATH is the absolute path to the root directory of dataset. DEEP_SPEECH_MODELS_PATH is the absolute path to Mozilla DeepSpeech's model folder.

python benchmark.py --dataset_type DATASET_TYPE --dataset_root DATASET_PATH --deep_speech_model_path DEEP_SPEECH_MODELS_PATH/output_graph.pbmm \
--deep_speech_alphabet_path DEEP_SPEECH_MODELS_PATH/alphabet.txt --deep_speech_language_model_path DEEP_SPEECH_MODELS_PATH/lm.binary \
--deep_speech_trie_path DEEP_SPEECH_MODELS_PATH/trie

The above prints the WER for different engines in console.

time command is used to measure execution time of different engines for a given audio file and then divide the CPU time by audio length. In order to measure execution time for Cheetah run

time resources/cheetah/cheetah_demo resources/cheetah/libpv_cheetah.so resources/cheetah/acoustic_model.pv \
resources/cheetah/language_model.pv resources/cheetah/cheetah_eval_linux_public.lic PATH_TO_WAV_FILE

The output should have the following format (values will be different)

real	0m4.961s
user	0m4.936s
sys	0m0.024s

then divide user by length of the audio file in seconds. The user is the actual CPU time spent in the program.

For DeepSpeech

time deepspeech --model DEEP_SPEECH_MODELS_PATH/output_graph.pbmm --alphabet PATH_TO_WAV_FILE DEEP_SPEECH_MODELS_PATH/alphabet.txt \
--lm DEEP_SPEECH_MODELS_PATH/lm.binary --trie DEEP_SPEECH_MODELS_PATH/trie --audio PATH_TO_WAV_FILE

Finally, for PocketSphinx

time pocketsphinx_continuous -infile PATH_TO_WAV_FILE

Valgrind's massif tool is used to measure heap memory usage. For example

valgrind --tool=massif pocketsphinx_continuous -infile PATH_TO_WAV_FILE

It creates a file with naming like massif.out.XXXX. The file can be read using

ms_print massif.out.XXXX

Below results are obtained by following the steps above. The benchmarking is performed on a laptop running Ubuntu 18.04 with 8 GB of RAM and Intel i7-4510U CPU running at 2GHz. WER refers to word error rate and RTF refers to real time factor.

Cheetah achieves higher accuracy compared to any other engine on both datasets. Compared to second best performing engine, Mozilla DeepSpeech, it is 13.9 times faster and consumes 21.4 times less memory. This enables Cheetah to run on small commodity embedded platforms such as Raspberry Pi while delivering the benefits of large models that need much more compute/memory resources.

The benchmarking framework is freely-available and can be used under the Apache 2.0 license. Regarding Mozilla DeepSpeech and PocketSphinx please refer to their respective pages.

The provided Cheetah resources (binary, model, and license file) are the property of Picovoice. They are only to be used for evaluation purposes and their use in any commercial product is strictly prohibited.

For commercial inquiries regarding Cheetah please contact us at [email protected]. For partnership opportunities contact us at [email protected].