mayamatakeshi/julius_server

This is a forking server for julius speech recognition engine. It waits on a TCP port for clients and forks processes to handle them.

julius_server's Issues

forked process must periodically try to write to the socket

We might add a msg like '+POLL'.
This is to ensure the other end is still alive so that we don't keep the app alive preventing new forks to happen (because of MAX_CONCURRENT_SESSIONS).

Implement speech timeout

First we need to specify a message to pass this value.
It could be be the first line in the TCP socket like this:

+RECOGNIZE:timeout=X

(we might add other parameters like language if we decide it to make a multi language daemon).

This might be doable by using

_CALLBACK_EVENT_SEGMENT_BEGIN | On short-pause segmentation and decoder-based VAD mode, this callback will be called at the beginning of each segment, segmented by short pauses.

CALLBACK_EVENT_SEGMENT_END | On short-pause segmentation and decoder-based VAD mode, this callback will be called at the end of each segment, segmented by short pauses._

https://www.sp.nitech.ac.jp/~ri/julius-dev/doxygen/julius/4.0/ja/callback_8h.html#06fc87d81c62e9abb8790b6e5713c55bf241f0a78f37c906f8836b02f6bcd4a5

Implement simultaneous clients limit

We should count calls to fork and SIGCHLD and prevent forking if there are more than XX operations already in progress.
If limit is reached, we should close the connection immediately so that some load balancer like haproxy/nginx can try another server.