whisper.cpp/examples/lsp/whisper.vim
AustinMroz 175ffa64ee
examples : vim plugin and LSP server (#1144)
* Initial proof of concept Vim plugin

At present, this is likely only slightly better than feature parity with
the existing whisper.nvim

Known issues:
 Trailing whitespace
 Up to an existing length(5 seconds) of speech may be processed when
  listening is enabled
 CPU cycles are spent processing speech even when not listening.

Fixing these issues is likely dependent upon future efforts to create a
dedicated library instead of wrapping examples/stream

* Support $WHISPER_CPP_HOME environment variable

A minor misunderstanding of the whisper.nvim implementation resulted in
a plugin that was functional, but not a drop in replacement as it should
be now.

* Initial progress on LSP implementation

Libcall is nonviable because the library is immediately freed after a
call is made. Further investigation has shown Language Server Protocol as
a promising alternative that both simplifies the required logic on the
vimscript side and increases the ease with which plugins for other
editors could be made in the future. This is a very large undertaking
and my progress has slowed substantially.

Work is far from being in a usable state, but I wish to keep track of
major refactors for organizational purposes.

* Rewrite audio windowing of guided transcription

One of the defining goals of this venture is allowing consecutive
commands to be rattled off without the existing deadzones of the current
implementation.

* Add unguided_transcription. Cleanup.

The unguided transcription implantation heavily borrows from existing
example implementations and the guided_transcription logic.

A high level pass was done to check that method arguments are accurate
to what inputs are actually required.

A first attempt at cancellation support was added for record keeping,
but will be deleted in a future commit.

* Fix compilation.

Resolves a large number of compilation errors.
No testing has been done yet for execution errors.

Update Makefile and .gitignore

* Functional unguided_transcription

* Functional guided_transcription

Fix commandset_list being passed by value
Properly register the first token of a multitoken command

* Minor changes before time fix

I've apparently made an awfully major mistake in thinking that unix time
was in milliseconds and will be changing all timekeeping code to use
standardized methods.

In preparation for this is a number of minor bugfixes.
Output is manually flushed.
An echo method has been added.
registerCommandset now wraps the returned index

* Swap timekeeping to use std::chrono

* Add work in progress lsp backed whisper.vim plugin

Current progress blockers are
 Adding modality awareness to the command processing
  (specifically, motion prompting)
 Improving the VAD to be a little more responsive
  (testing start of activity)

* Reworked vim plugin command loop

* Fix change inside

Multiple bug fixes that, crucially, bring the plugin to the point where a
demonstration video is possible

Add better echo messaging so whisper_log isn't required
 Add loading complete message as indicator when listening has started
Insert/append are actually included in command sets
Some more heavy handed corrections to prevent a double exit when leaving
insert mode
As a somewhat hacky fix, the very first space is removed when inserting.
 This cleans up most use cases, but leaves me unsatisfied with the few
 cases it would be desired.

* Forcibly set commandset_index to 0 after subinsert

Also remove unnecessary ! to use builtin vim command

* Fix upper

A minor scope mistake was causing upper'd inputs to be eaten.
This was fixed and echoing was slightly improved for clarity.

* Fix formatting

Corrects indentation to 4 spaces as project standard
Slightly better error support for malformed json input

* Remove obsolete vim plugin

* Add json.hpp library

The same library that is used for the llama.cpp server

* Minor cleanups

add lsp to the make clean directive.
remove a redundant params definition.
reorder whisper.vim logging for subtranscriptions
Corrections to unlets (variables of argument scope appear immutable)

* Fix indentation. Fallback for subTranscription

Indentation has been changed to 4 spaces.

Unit testing has been set up, I'm opting not to include it in the
repository for now.
It however, has revealed a bug in the state logic where a
subtranscription can be initiated without having a saved command
When this occurs, append is added as a fallback

* Move audio polling logic to a subfunction

While work on the improved vad will continue, It's grown to be a little
out of scope. Instead, a future commit will perform multiple detection
passes at substretches of audio when a backlog of audio exists.

To facilitate this, and prevent code duplication, the vad code has been
moved into a subfunction shared by both the unguided and guided
transcription functions.

* Test for voice over subchunks if backlog > 1s

As the existing VAD implementation only checks for a falling edge at the
end of an audio chunk. It fails to detect voice in cases where the
recorded voice is only at the beginning of the audio.

To ameliorate this, when the timestamp would cause analysis of audio
over a second in length, it is split into 1 second length subchunks
which are individually tested.

Results are promising, but there seems to be a remaining bug with
unguided transcription likely related to saving context

* Limit the maximum length of audio input.

This existing VAD implementation only detects falling edges, which
means any gap in the users speaking is processed for transcription.
This simply establishes a constant maximum length depending on the type
of transcription. Uguided gets a generous 10 seconds and guided, 2.

While quick testing showed that commands are generally around a half a
second to a second, limiting commands to an even second resulted in
extreme degradation of quality. (Seemingly always the same output for a
given commandset)

* Unguided timestamp tracking, cleanup

Unguided transcriptions where not setup to allow for passing of
timestamp data forward, but have been corrected.

No_context is now always set to false. While conceptually desirable for
the quality of guided transcription, It was seemingly responsible for
prior command inputs ghosting in unguided transcription.

Save and Run are now tracked by command number instead of command text.
While command_text was provided for convenience, I wish to keep command
index authoritative. This gives greater consistency and potentially
allows for end users to rename or even translate the spoken versions of
these commands

* By default, maintain mode.

Previously, mode was reset to 0 unless otherwise set.
In addition to causing some edge cases, this was didn't mesh well with
the existing approach to visual mode.

With this change, initial tests indicate visual mode is functional.

* Add undo breaks before subtranscriptions

Subtranscriptions use undo as a hack to allow for partial responses to
be displayed. However, scripts don't cause an undo break mid execution
unless specifically instructed to. This meant that multiple
unguided transcriptions from a single session would cause a latter to
undo a former.

This is now fixed and undo should be reasonably usable as a command.

* Append instead of insert for new undo sequence

When entering and leavening insert mode with `i`, the cursor shifts one
column to the left. This is remedied by using append instead of insert
for setting these breaks in the undo sequence

`-` was also added to the pronunciation dictionary to be pronounced as
minus as it was causing a particularly high failure rate.

* Move undo sequence breaks to command execution

Previously, undo sequence breaks were triggered when there was a command
that caused a move to insert mode. This caused commands that changed
state (like delete or paste) to be bundled together with into the last
command that caused text to be entered.

* Fix repeat. Add space, carrot, dollar commands

 Repeat (.) wasn't being tracked properly just like undo and is being
 manually tracked now.

 While efforts have been made to properly handle spaces, it was
 particularly finicky to add a single space when one is needed. A
 special 'space' command has been added to insert a single space and move
 the cursor after it.

 Carrot and Dollar commands have been added for start of line and end of
 line respectively. These are both simple to implement, and just a
 matter of defining a pronunciation.

* Return error on duplicate in commandset

Not every command in the commandset tokenizes to a single token.
Because of this, it's possible for that two commands could resolve to
the same single token after subsequent tokens are discarded.

This commit adds a simple check for duplicates when a commandset is
registered and returns an error if so.

Additional code will be required later on the vim side to actually
process this error.

* Add support for user-defined commands

This adds a user definable dictionary from spoken keys to strings or
funcrefs. All keys are added to the commandlist and when spoken, trigger
the corresponding function.

Like "save" and "run", these user commands are only available when the
command buffer is empty.

* Add readme, update cmake

* Add area commandset. Refactor spoken_dict

Area commands (inside word, around sentence...) have been given a
commandset as considered earlier.

Verbose definitions for spoken_dict entries now use dicts instead of
lists. This shortens the definition for most keys that require it and
scales better with the addition of further commandsets

* Add mark, jump. Fix change under visual.

Mark (m) and jump (') have been added.

When a visual selection was executed upon a command that initiated a
subtranscription (change) the area of the visual selection is not
properly tracked which causes the attempt to stream in partial response
to fail. This is solved by disabling partial transcriptions from being
streamed when a subtranscription is started while in visual mode.

* Accommodate ignorecase. Fix change.

From testing on older different versions of vim, the test for
distinguishing an 'R' replace all from an 'r' replace could fail if
ignorecase was set. The comparison has been changed to explicitly
require case matching

Change detection has been moved to the execution section as it was missing the
change+motion case.

* Support registers. Fix README typo

There's no logic to prevent doubled register entry, but the functional
result is equivalent to if the same key order was typed into vim.

A minor typo in the readme. I've mismemorized the mnemonic for 't' as 'to'
instead of till., but 'to' can't be used as it's a homophone with '2'.
While there was no mistake in the actual logic, it was misleading to use
'to' in the readme.
2023-08-27 21:35:06 +03:00

363 lines
16 KiB
VimL

if !exists("g:whisper_dir")
let g:whisper_dir = expand($WHISPER_CPP_HOME)
if g:whisper_dir == ""
echoerr "Please provide a path to the whisper.cpp repo in either the $WHISPER_CPP_HOME environment variable, or g:whisper_dir"
endif
endif
if !exists("g:whisper_lsp_path")
let g:whisper_lsp_path = g:whisper_dir .. "lsp"
if !filereadable(g:whisper_lsp_path)
echoerr "Was not able to locate a lsp executable at: " .. g:whisper_lsp_path
throw "Executable not found"
endif
endif
if !exists("g:whisper_model_path")
" TODO: allow custom paths relative to the repo dir
let g:whisper_model_path = g:whisper_dir .. "models/ggml-base.en.bin"
if !filereadable(g:whisper_model_path)
echoerr "Could not find model at: " .. g:whisper_model_path
throw "Model not found"
endif
endif
let s:output_buffer = bufnr("whisper_log", v:true)
call setbufvar(s:output_buffer,"&buftype","nofile")
let s:lsp_command = [g:whisper_lsp_path,"-m",g:whisper_model_path]
" For faster execution. TODO: server load multiple models/run multiple servers?
" let s:lsp_command = [g:whisper_lsp_path, "-m", g:whisper_dir .. "models/ggml-tiny.en.bin", "-ac", "128"]
" requestCommands([params_dict])
func whisper#requestCommands(...)
let l:req = {"method": "guided", "params": {"commandset_index": 0}}
if a:0 > 0
call extend(l:req.params, a:1)
endif
let resp = ch_sendexpr(g:lsp_job, l:req, {"callback": function("s:commandCallback", [l:req.params, 0])})
endfunction
" doTranscription([params_dict])
func whisper#doTranscription(...)
let l:req = {"method": "unguided", "params": {}}
if a:0 > 0
call extend(l:req.params, a:1)
endif
let resp = ch_sendexpr(g:lsp_job, l:req, {"callback": function("s:transcriptionCallback", [function("s:insertText"),function("s:endTranscription")])})
endfunction
" For testing
func whisper#uppertest(cha)
echo tr(a:cha, s:c_lowerkeys, s:c_upperkeys)
endfunction
" (upper, exit, count, motion, command, insert/append, save run) "base"
" (upper, exit, count, motion, command, inside/around) "motion/visual"
" (upper, exit, count, motion, line, inside/around) "command already entered"
" (upper, exit, key, ) "from/till"
" upper and lower keys is used to translate between cases with tr
" Must be sunchronized
let s:c_lowerkeys = "1234567890-=qwertyuiop[]\\asdfghjkl;'zxcvbnm,./\""
let s:c_upperkeys = "!@#$%^&*()_+QWERTYUIOP{}|ASDFGHJKL:\"ZXCVBNM<>?'"
let s:c_count = split("1234567890\"",'\zs')
let s:c_command = split("ryuogpdxcv.iam", '\zs')
let s:c_motion = split("wetf'hjklnb$^)",'\zs')
" object words: Word, Sentence, Paragraph, [, (, <, Tag, {. ", '
let s:c_area = split("wsp])>t}\"'",'\zs')
"Special commands.
let s:c_special_always = ["exit", "upper"]
let s:c_special_normal = ["save", "run", "space"]
" If not in dict, key is spoken word,
" If key resolves to string, value is used for normal/motion, but key for chars
" If key resolves to dict, {0: "normal",1: "motion",2:"single char",3: "area"}
" Missing entries fall back as follows {0: "required", 1: 0, 2: "key", 3: 0}
let s:spoken_dict = {"w": "word", "e": "end", "r": "replace", "t": {0: "till", 3: "tag"}, "y": "yank", "u": "undo", "i": {0: "insert", 1: "inside"}, "o": "open", "p": {0: "paste", 3: "paragraph"}, "a": {0: "append", 1: "around"}, "s": {0: "substitute", 3: "sentence"}, "d": "delete", "f": "from", "g": "go", "h": "left", "j": "down", "k": "up", "l": "right", "c": "change", "v": "visual", "b": "back", "n": "next", "m": "mark", ".": {0: "repeat", 2: "period"}, "]": {0: "bracket", 2: "bracket"}, "'": {0: "jump", 2: "apostrophe", 3: "apostrophe"}, '"': {0: 'register', 2: "quotation", 3: "quotation"}, "-": {0: "minus", 2: "minus"}, "$": {0: "dollar", 2: "dollar"}, "^": {0: "carrot", 2: "carrot"}, ")": {0: "sentence", 2: "parenthesis", 3: "parenthesis"}, "}": {0: "paragraph", 2: "brace", 3: "brace"}, ">": {0: "indent", 2: "angle", 3: "angle"}}
" Give this another pass. This seems overly hacky even if it's functional
let s:sub_tran_msg = ""
func s:subTranProg(msg)
if s:sub_tran_msg != ""
let s:sub_tran_msg = s:sub_tran_msg .. a:msg
if mode() !=? 'v'
exe "normal" "u" .. s:sub_tran_msg
endif
else
if s:command_backlog == ""
" this should not occur
call s:logCallback(0, "Warning: Encountered sub transcription without prior command")
let s:command_backlog = "a"
endif
if a:msg[0] == ' '
let s:sub_tran_msg = s:command_backlog .. a:msg[1:-1]
else
let s:sub_tran_msg = s:command_backlog .. a:msg
endif
if mode() !=? 'v'
exe "normal" s:sub_tran_msg
endif
endif
call appendbufline(s:output_buffer, "$", s:sub_tran_msg .. ":" .. string(a:msg ))
endfunction
func s:subTranFinish(params, timestamp)
let s:repeat_command = s:sub_tran_msg
" Visual selection is lot if used with streaming, so streaming of partial
" transcriptions is disabled in visual mode
if mode() ==? 'v'
exe "normal" s:sub_tran_msg
endif
let s:sub_tran_msg = ""
let s:command_backlog = ""
exe "normal a\<C-G>u"
let l:params = a:params
let l:params.timestamp = a:timestamp
if exists("l:params.commandset_index")
unlet l:params.commandset_index
endif
call whisper#requestCommands(a:params)
endfunction
func s:logCallback(channel, msg)
call appendbufline(s:output_buffer,"$",a:msg)
endfunction
func s:transcriptionCallback(progressCallback, finishedCallback, channel, msg)
let l:tr = a:msg.result.transcription
let l:ex_ind = match(tolower(l:tr),"exit", len(l:tr)-6)
" The worst case I've observed so far is " Exit.", which is 6 characters
if l:ex_ind != -1
call a:progressCallback(strpart(l:tr,0,l:ex_ind-1))
call a:finishedCallback(a:msg.result.timestamp)
else
call a:progressCallback(l:tr)
let req = {"method": "unguided", "params": {"timestamp": a:msg.result.timestamp, "no_context": v:true}}
let resp = ch_sendexpr(g:lsp_job, req, {"callback": function("s:transcriptionCallback", [a:progressCallback, a:finishedCallback])})
endif
endfunc
func s:insertText(msg)
exe "normal a" .. a:msg
endfunction
func s:endTranscription(timestamp)
call appendbufline(s:output_buffer, "$", "Ending unguided transcription")
endfunction
" If a command does not include a whole actionable step, attempting to execute
" it discards the remainder of things. There is likely a simpler solution,
" but it can be made functional now by storing a backbuffer until actionable
let s:command_backlog = ""
let s:repeat_command = ""
let s:preceeding_upper = v:false
func s:commandCallback(params, commandset_index, channel, msg)
let l:command_index = a:msg.result.command_index
let l:do_execute = v:false
let l:next_mode = a:commandset_index
let l:command = s:commandset_list[a:commandset_index][l:command_index]
call s:logCallback(0, string(a:msg) .. " " .. a:commandset_index .. " " .. l:command)
if l:command_index == 0
"exit
"if s:command_backlog == ""
call s:logCallback(0,"Stopping command mode")
echo "No longer listening"
let s:command_backlog = ""
return
"else
" Legacy code to clear an existing buffer with exit.
" Was found to be rarely desired and is better introduced as a
" standalone command (clear?)
" call s:logCallback(0,"Clearing command_backlog" .. s:command_backlog)
" let s:command_backlog = ""
" let s:preceeding_upper = v:false
" endif
elseif l:command_index == 1
" upper
let s:preceeding_upper = !s:preceeding_upper
elseif l:command == "save"
" save and run can only happen in commandset 0,
exe "w"
elseif l:command == "run"
exe "make run"
elseif l:command == "space"
exe "normal i \<ESC>l"
elseif has_key(s:c_user, l:command)
let Userfunc = s:c_user[l:command]
if type(Userfunc) == v:t_string
let Userfunc = function(Userfunc)
endif
call Userfunc()
else
if s:preceeding_upper
" Upper should keep commandset
let s:preceeding_upper = v:false
let l:visual_command = tr(l:command, s:c_lowerkeys, s:c_upperkeys)
else
let l:visual_command = l:command
endif
echo s:command_backlog .. " - " .. l:visual_command
let s:command_backlog = s:command_backlog .. l:visual_command
if a:commandset_index == 2 || a:commandset_index == 3
" single key, either completes motion, replace, or register
" Should move to execute unless part of a register
" Change will be caught at execute
if s:command_backlog[-2:-2] !=# '"'
call s:logCallback(0,"not register")
let l:do_execute = v:true
end
let l:next_mode = 0
" commandset index only matters for a/i
elseif (l:command == "a" || l:command == "i") && a:commandset_index == 1
" inside/around. Is commandset 3
let l:next_mode = 3
elseif l:command ==# '"'
let l:next_mode = 2
elseif index(s:c_count, l:command) != -1
let l:next_mode = a:commandset_index
elseif index(s:c_motion, l:command) != -1
if l:command == 't' || l:command == 'f' || l:command == "'"
" prompt single key
let l:next_mode = 2
else
let l:do_execute = v:true
let l:next_mode = 0
endif
elseif index(s:c_command, l:command) != -1
if index(["y","g","d","c"], s:command_backlog[-1:-1]) != -1 && s:command_backlog[-1:-1] != s:command_backlog[-2:-2] && mode() !=? 'v'
" need motion or repeated command
" Potential for bad state here if disparaging command keys are
" entered (i.e. yd), but vim can handle checks for this at exe
" And checking for cases like y123d would complicate things
let l:next_mode = 1
elseif index(["i","a","c", "o", "s"], l:command) != -1 || s:command_backlog[-1:-1] ==# 'R'
"'Insert' mode, do general transcription
let l:req = {"method": "unguided", "params": a:params}
let l:req.params.timestamp = a:msg.result.timestamp
let l:req.params.no_context = v:true
let resp = ch_sendexpr(g:lsp_job, req, {"callback": function("s:transcriptionCallback", [function("s:subTranProg"), function("s:subTranFinish", [a:params])])})
return
elseif l:command == 'r' || l:command == 'm'
let l:next_mode = 2
elseif l:command == '.'
let l:next_mode = 0
let l:do_execute = v:true
let s:command_backlog = s:command_backlog[0:-2] .. s:repeat_command
else
if l:command ==? 'v'
let l:next_mode = 1
else
let l:next_mode = 0
endif
let l:do_execute = v:true
endif
else
throw "Invalid command state: " .. l:command .. " " .. a:commandset_index .. " " .. s:command_backlog
endif
endif
if l:do_execute
if mode() ==?'v' && l:next_mode == 0
let l:next_mode = 1
elseif match(s:command_backlog, 'c') != -1
let l:req = {"method": "unguided", "params": a:params}
let l:req.params.timestamp = a:msg.result.timestamp
let l:req.params.no_context = v:true
let resp = ch_sendexpr(g:lsp_job, req, {"callback": function("s:transcriptionCallback", [function("s:subTranProg"), function("s:subTranFinish", [a:params])])})
return
endif
exe "normal" s:command_backlog
if index(s:c_motion + ["u"],l:command) == -1
exe "normal a\<C-G>u"
let s:repeat_command = s:command_backlog
call s:logCallback(0, s:command_backlog)
endif
let s:command_backlog = ""
endif
let l:req = {"method": "guided", "params": a:params}
let l:req.params.timestamp = a:msg.result.timestamp
let l:req.params.commandset_index = l:next_mode
let resp = ch_sendexpr(g:lsp_job, l:req, {"callback": function("s:commandCallback",[a:params, l:next_mode])})
endfunction
func s:loadedCallback(channel, msg)
echo "Loading complete"
call s:logCallback(a:channel, a:msg)
endfunction
func s:registerCommandset(commandlist, is_final)
let req = {"method": "registerCommandset"}
let req.params = a:commandlist
call s:logCallback(0, join(a:commandlist))
call add(g:whisper_commandlist_spoken, a:commandlist)
if a:is_final
let resp = ch_sendexpr(g:lsp_job, req, {"callback": "s:loadedCallback"})
else
let resp = ch_sendexpr(g:lsp_job, req, {"callback": "s:logCallback"})
endif
endfunction
func s:registerAllCommands()
let l:normal = s:c_special_always + s:c_special_normal + s:c_count + s:c_command + s:c_motion + keys(s:c_user)
let l:visual = s:c_special_always + s:c_count + s:c_command + s:c_motion
" Currently the same as visual.
" let l:post_command = s:c_special_always + s:c_count + s:c_command + s:c_motion
let l:single_key = s:c_special_always + split(s:c_lowerkeys, '\zs')
let l:area = s:c_special_always + s:c_area
" Used only for compatibility with the testing script
let g:whisper_commandlist_spoken = []
let s:commandset_list = [l:normal, l:visual, l:single_key, l:area]
call s:registerCommandset(s:commandsetToSpoken(l:normal, 0), v:false)
call s:registerCommandset(s:commandsetToSpoken(l:visual, 1), v:false)
call s:registerCommandset(s:commandsetToSpoken(l:single_key, 2), v:false)
call s:registerCommandset(s:commandsetToSpoken(l:area, 3), v:true)
endfunction
func s:commandsetToSpoken(commandset, spoken_index)
let l:spoken_list = []
for l:command in a:commandset
if has_key(s:spoken_dict, l:command)
let l:spoken_value = s:spoken_dict[l:command]
if type(l:spoken_value) == v:t_dict
if has_key(l:spoken_value, a:spoken_index)
let l:spoken_value = l:spoken_value[a:spoken_index]
else
if a:spoken_index == 2
let l:spoken_value = l:command
else
let l:spoken_value = l:spoken_value[0]
endif
endif
else
if a:spoken_index == 2
let l:spoken_value = l:command
endif
endif
else
let l:spoken_value = l:command
endif
call add(l:spoken_list, l:spoken_value)
endfor
return l:spoken_list
endfunction
" TODO: Check lifetime. If the script is resourced, is the existing
" s:lsp_job dropped and therefore killed?
" This seems to not be the case and I've had to deal with zombie processes
" that survive exiting vim, even though said behavior conflicts with my
" understanding of the provided documentation
let s:lsp_opts = {"in_mode": "lsp", "out_mode": "lsp", "err_mode": "nl", "err_io": "buffer", "err_buf": s:output_buffer}
if !exists("g:lsp_job")
if exists("g:whisper_user_commands")
let s:c_user = g:whisper_user_commands
else
let s:c_user = {}
endif
let g:lsp_job = job_start(s:lsp_command, s:lsp_opts)
if job_status(g:lsp_job) == "fail"
echoerr "Failed to start whisper job"
endif
call s:registerAllCommands()
endif