Projects
From SpeechWiki
Here are some projects that SST People are working on. For another view, see our Publications.
Contents |
SST Group Meetings
Phonetics, Phonology, Semantics
- Prosody and Phonology in Automatic Speech Recognition (Landmark-Based Speech Recognition)
- Group Meeting Schedules and Slides
- Landmark-Based Speech Recognition
- Prosody of Disfluency
- Very Large Corpus ASR/ Mixed-Units ASR
- Large Vocabulary speech recognition using mixed units on fisher corpus
- Articulatory Feature Transcription
- Transcription Guidelines
- Phone-to-Feature Mapping
- Meeting Summaries
- Resources
Group dynamics and Discourse
- GroupScope --- Dynamics of Medium-Sized Groups
- GroupScope
Language Acquisition, Language Contact, Variability, and Disability
- Multi-Dialect Speech Recognition and Machine Translation for Qatari Broadcast TV
- Multi Dialect Arabic
- Cross-Language Transfer Learning
- Linguistic Diversity References
- Star Challenge competition
- Dynamics of Second Language Fluency
- Group Meeting Schedules and Slides
- Description
- Data Description
- Universal Access
- Group Meeting Schedules and Slides
- Description
- UASpeech Database
Multimodal Fusion, Speech and Non-Speech
- Audiovisual Event Detection and Visualization
- Group Meeting Schedules and Slides
- Papers
- Visualization Experiments
- Mobile Platform Acoustic-Frequency Environmental Tomography (was Dereverberation)
- Group Meeting Schedules
- Project Status and Working Notes
- Audiovisual Speech Recognition
- Description
- AVICAR Database
- Smaragdis collaboration
-
-
Pseudocode spec for the sound input class (and also output later, but not read-and-write):
class input_t{ // Definition of stream characteristics class specs_t{ size_t channels; double sample_rate; enum sample_format; };
// // Constructors //
input_t( ??? stream, bool in_or_out, size_t ch, double sr, enum frm) { switch( stream){ case "file" use ffmpeg case "socket" use homebrew code? case "url" use VLC? case "adc" Use portaudio case "dac" Use portaudio } }
input_t( ??? stream, input_t example); // copy stream attributes input_t( ??? stream, input_t::specs_t example); // copy stream attributes
Assignment/copy operators
// // Destructor //
~input_t() // bookkeeping with closing file/net/etc.
// // Utilities //
double sample_rate(); size_t channels(); enum sample_format(); bool eof(); bool();
// // Seeking //
seek( size_t s); // move to sample frame s seek( double t); // move to second t
// // Reading // output should be channels by sample frames
array<T> &read( size_t n, size_t offset, int channel_mask); // sample frames array<T> &read( double n, double offset, int channel_mask); // seconds
// // Writing //
write( array<T> &x, size_t offset, int channel_mask); // sample frames write( array<T> &x, double offset, int channel_mask); // seconds
write_add( array<T> &x, size_t offset, int channel_mask); // sample frames write_add( array<T> &x, double offset, int channel_mask); // seconds };
We are going for a blocking interface instead of cumbersome callbacks for now. The stream parameters when reading can be used to perform +on the fly resampling and channel remapping. I'm attaching the board doodling in case I missed something.
We are currently working on the getting code to work for the simple case:
main() { input_t in( ...);
while( in){ x = in.read( ...); y = feature( x); plot( y); } }
I'm working on the feature object, Camille is working on the input object.
