6a7c82501e
whisper : improve decoding strategy ( #244 )
...
- Clear past prompt when there is very short audio left for processing.
My observation is that in these cases the decoding tends to repeat and
hallucinate stuff and I think this is induced by the existing prompt
- When we fail to sample timestamp token, retry by clearing the past
prompt. If it fails again, then we advance the window by 1 second
2 years ago
a82d331034
stream : update README.md + comments
2 years ago
c37c2443c1
Update README.md ( #56 )
2 years ago
0f11759406
ggml : make more compatible with c99 ( #262 )
2 years ago
5a5c5ddcca
Update README.md
2 years ago
34e0b4b9ef
stream : fix build
2 years ago
b0f8013eb9
stream : add sliding window mode
2 years ago
124c718c73
whisper : fix UB when reading buffer of length 0 bytes ( #265 )
2 years ago
f66ac6dc4f
ggml : fix indentation
2 years ago
9955fa4ed7
ggml : make compatible with c99 ( #262 )
2 years ago
a613f16aec
talk : improve prompting
2 years ago
930c693989
release : v1.0.3
...
Fixed whisper.spm tests
2 years ago
d8a0dde31a
Update README.md
2 years ago
9e3e6f253a
release : v1.0.2
2 years ago
57ccd7cc4f
Update README.md
2 years ago
812ae3ffbd
Update README.md
2 years ago
f309f97df6
Node.js package ( #260 )
...
* npm : preparing infra for node package
* npm : package infra ready
* npm : initial version ready
* npm : change name to whisper.cpp
whisper.js is taken
2 years ago
aa6adda26e
talk : make compatible with c++11 (part 2)
2 years ago
444349f4ec
talk : make compatible with c++11
2 years ago
37a93d2459
cmake : require c++11 instead of c++20
2 years ago
e70d47baab
Remove C++20 requirement ( #257 )
...
* Remove C++20 requirement
* Roll back C features not supported in VS2017
2 years ago
6ed786957e
Add newline per segment for text output ( #254 )
2 years ago
ea38ad6e70
bench : more concise representation of the results ( #89 )
2 years ago
054940e1f6
minor : fix .gitignore to not ignore examples
2 years ago
fcf515de60
bench.wasm : same as "bench" but runs in the browser ( #89 )
2 years ago
85c9ac18b5
Update README.md
2 years ago
b7c85d1ea6
talk : fix build for MSVC
2 years ago
3b1aacbe6d
talk : talk with AI in the terminal
2 years ago
d1da35de06
fix potential bug reading model data into a small size optimized string which could lead to memory corruption. In an SSO string, you can't write data to &str[0] and expect it to work well.
...
Also added a small wrapper function to more safely read model data without having to get the sizeof right. I tested this on tiny, base and large models, there was no change in behaviour.
2 years ago
603f97ba11
whisper : minor improvemnt in decoding strategy ( #244 )
...
Do not allow for text segments to go beyond end of audio.
This partially mitigates some issues when the last audio window is 1-2
seconds just before the end of the audio file and the decoding spirals
into a repetition of the last transcribed phrase.
2 years ago
50a061b313
ggml : add alternative cblas_sgemm call
2 years ago
832b4f34c9
make : indentation + .gitignore
2 years ago
0f98755fc5
Flag for Position Independent Code
2 years ago
56822621a8
twitch.sh : various fixes and polishing
...
- check if streamlink is installed
- fix audio chunking
- change default threads to 4
2 years ago
9e5f3ddc16
Allow for Twitch.tv live transcription
...
We rely on streamlink library to give us a stream, then we proceed similarly to
the radio livestream example.
2 years ago
d91c001120
Fix paths echoed after the download
...
Was using models path instead of root path
2 years ago
04a16bbf11
fix compilation on haiku
2 years ago
47afb93c3c
yt-wsp.sh : improve usage instructions
2 years ago
575c53dc41
yt-wsp.sh : fix usage instruction + comment
2 years ago
3996ecc156
Update README.md
2 years ago
faa85f9840
livestream.sh : remove obsolete comment
2 years ago
b6597539f9
ggml : fix typo in previous commit
2 years ago
9a4b7a916e
ggml : use macros to inline FP16 <-> FP32 conversions
2 years ago
f8ec718b76
ggml : add F16C CPU flag check
2 years ago
35b40a93b9
add fp16/fp32 convert intrinsics
2 years ago
9fe7306f4b
models : add the new "large" model release by OpenAI
...
The old "large" model is now renamed "large-v1".
If you have been using it, make sure to rename it and download the new
"large" model for best results.
2 years ago
13e8eb2346
bench : add commit hash to bench-all.sh results
2 years ago
78d13257be
Try to improve the token sampling strategy ( #193 )
...
* whisper : try to improve the token sampling strategy
- Add the "max_initial_timestaamp" token logic from OpenAI
- Disallow sampling timestamps that are in the past
* whisper : fix the max initial timestamp logic + fallback decoding
2 years ago
9b7df68753
tests : adding transcription tests
2 years ago
061fc81bd6
ggml : remove inline specifier from fp16 <-> fp32 converters
2 years ago
Georgi Gerganov
Roland Rabien
Lexevolution
bert hubert
Reinis Muiznieks
keyehzy
Kartik Saranathan
Al Hoang
katsu560