Sample rate for Russian acoustic model

[Dmitry Suvorov]

Hello!

On what sample rate is trained Russian acoustic model available in archive cmusphinx-ru-5.2.tar.gz which was uploaded on 28 of April? I didn't find any config file where it is written.

[Nickolay V. Shmyrev]

16

[Dmitry Suvorov]

Thank you for your quick answer, Nickolay!

I downloaded the latest version of pocketsphinx and sphinxbase from git repository and tried to use it with this model using C API. Code for initialization is following:

m_config = cmd_ln_init( nullptr, ps_args( ), TRUE,
                           "-hmm", "/home/ubuntu/cmusphinx-ru-5.2",
                           "-lm", "/home/ubuntu/cmusphinx-ru-5.2/ru.lm",
                           "-dict", "/home/ubuntu/cmusphinx-ru-5.2/ru.dic",
                           nullptr );
    if( !m_config ) {
        throw std::runtime_error( "Failed to create config object, see log for details" );
    }
    m_ps = ps_init( m_config );
    if( !m_ps ) {
        throw std::runtime_error( "Failed to create recognizer, see log for details" );
    }

But I got error for each word in the dictionary like following:
ERROR: "dict.c", line 195: Line 65914: Phone 'е1' is mising in the acoustic model; word 'встрепанная' ignored
The full log is in attachments. What am I doing wrong?

[Nickolay V. Shmyrev]

It is a bug in a model, download it again and retry.

[Dmitry Suvorov]

Now system started to work. But I always get error after a random number of phrases are recognized:

acmod.c:789: acmod_process_cep: Assertion `acmod->n_feat_frame <= acmod->n_feat_alloc' failed.

I checked using gdb that n_feat_frame is set to 15 and n_feat_alloc is set to 12. Nickolay, can you help me to solve this problem?

[Dmitry Suvorov]

I send samples to the decoder by 160 or 5600 units. Is there any limitations on number of samples sending to the decoder?
Also is there a way to compile sources in release mode (without asserts and debug info)?

[Nickolay V. Shmyrev]

But I always get error after a random number of phrases are recognized:

You need to provide more details on how to reproduce this issue in order to get help on the problem

I send samples to the decoder by 160 or 5600 units. Is there any limitations on number of samples sending to the decoder?

No

Also is there a way to compile sources in release mode (without asserts and debug info)?

No, logging must be disabled in runtime.