Go to the first, previous, next, last section, table of contents.

CHATR Commands

This appendix contains a list of all commands currently defined in CHATR, and a short description of their function. The list is automatically generated from the actual code definitions of the functions, in the hope that this will keep it up to date.

(+ NUM0 NUM1 ...)

Return the sum of the numbers.

(- NUM0 NUM1 ...)

Return the first number minus the rest.

(* NUM0 NUM1 ...)

Return the product of the numbers.

(/ NUM0 NUM1 ...)

Return value of the first divided by each of the following in turn.

(and CONJ0 ...)

Evaluate conjunctions until one evaluates to nil. Return nil or value of last conjunct.

(append LIST1 LIST2)

Return a list constructed by appending LIST1 LIST2 (destructively).

(apply FUNC ARGSLIST)

Apply FUNC to ARGSLIST and return the return value.

(assoc KEY ALIST)

Search for sublist in ALIST whose car is KEY and return that sublist.

(Audio COMMAND VALUE)

Set up audio output parameters. COMMAND may be one of

Device

Set type of hardware device. The device_type VALUE may be

SUN_AU: The Sun audio device (/dev/audio) 8KHz ulaw.
SUN_LIN16: The Sun audio device (/dev/audio) 16KHz shorts.
AF_CONN: AudioFile link (currently only 8k ulaw).
NA_CONN: Netaudio link.
DATLINK: Direct DATLINK connection.
BSD_SOCK: BSD Socket output.
NO_PLAY: Don't actually play anything.
AU_COMMAND: User specified command (see below).

Mode [Async | Sync]

In synchronous mode (default) CHATR wait for the waveform to complete playing before continuing. In asynchronous mode CHATR send the play request to the audio spooler daemon and continues execution immediately. Changing Async to Sync mode waits until the spool has less than one item.

Abort

In Async mode close down auspool NOW!

Required_Rate [None | SR ]

Force audio output to be resampled at 'SR' Hz. Only used if running CHATR in AU_COMMAND mode. If 'None' (default), output sample rate is as generated by synthesizer.

Required_Form [ulaw | lin16 | lin16MSB | lin16LSB]

Force audio output to be in desired format.

DataBase_Form [lin16MSB | lin16LSB]

Voice database format (default = speaker_default's format).

Resample_Quality [0 | 1 | 2]

Change parameters in resampling function. 0 is best (default) quality but slowest. 2 is worst quality but fastest.

Host

Host name or number. (Used for BSD_SOCK mode.)

Port

Port number to send data to on host.

Command

If Device is set to AU_COMMAND, VALUE is used as a UNIX command to play the waveform. The string may use the shell variables $FILE and $SR to refer to the raw data file (in native byte order) name and sample rate of the waveform to be played.

(boundp VARNAME)

Return 't' if this variable is already bound, nil otherwise.

(car LIST)

Return the 'car' of the given list.

(cd DIRNAME)

Change directory. With no argument, change to home directory.

(cdr LIST)

Return the 'cdr' of the given list.

(Cep_Filetype FILETYPE)

The default type for saving cepstrum parameter files. In unheadered case, users will have to determine the dimensions when reading them. All data is saved as floats. Be careful about byte order. FILETYPE may be

RAW: Raw.
HTK: HTK 12 byte header.
NUUTALK: The same as raw.
NIST: NIST header (not implemented yet).
ESPS: ESPS header (not implemented yet).

(dump_chatr FILENAME)

Dump an executable version of a CHATR binary into FILENAME using unexec. On restarting that binary, execution will continue from that point immediately after this call (though all files will be closed).

(Compare_Cepstrums UTT1 FILEID)

Find cepstral distances of synthesized unit from cepstrum vector of natural unit. A list of three values is returned, mean error, RMS error, and standard deviation of mean error. UTT1 must be an utterance with a cepstrum stream, i.e. it must have been synthesized using a synthesis method that generates a cepstrum description (e.g. NUUCEP). FILEID is an ID that when expanded with the CoefFileSkeleton field of the current database, is a path name to the cepstrum vectors of the target utterance. For example, for RADIO f2b, FILEID would be something like "st01/f2bst01p1", while in the Sally 200 database it would be "sc004".

(Compare_Contour UTT1 UTT2)

Compare the f0 contours of two generated utterances. The utterances should be of similar length.

(Compare_Int UTT1 UTT2)

Compare the intonation features between two utterances. The utterances should be of the same type of intonation features (tilt).

(cons OBJ1 OBJ2)

Return a new 'cons' cell whose 'car' is OBJ1 and 'cdr' is OBJ2.

(copy OBJ)

Return a copy of OBJ. Copies lists, strings, atoms, floats and ints. It does not go inside complex objects, such as streams, utterances, functions or pointers.

(Database COMMAND PARAMETERS)

Command to build, load and change speech unit databases. COMMAND may be one of

Set

Set a value in the current udb. PARAMETERS may be one of

Name

Name for the current udb.

IndexFile

File containing compiled unit index.

WaveFileSkeleton

String containing %%s.

PitchMarkFileSkeleton

String containing %%s

PitchMarkType

Type of pitchmarks (either preloaded or from files).

ALL_MARKED: Pitchmarks exist at regular intervals through even unvoiced sections (from fz_track).
VOICED_ONLY: Pitchmarks only in voicing (typically from get_f0 or egg).

PhoneSet

Phoneme set name.

Strategy

Generic, Simple, Hand, ... (others to follow).

UnitType

(int str ...) unit description.

WaveFileType

nist, ulaw (implies 8k), esps, raw (requires WaveSampleRate and WaveEncoding).

WaveSampleRate

In Hz.

WaveEncoding

lin16MSB, lin16LSB, lin16, ulaw.

Weights [CLEAR | weights]

Set weights for this database. If CLEAR then remove current weights and use weights as set by nus_params (old method). Otherwise use weights of form

     ( ( (ph0 ph1 ...) 
         (weightname0 weightval0)
         (weightname1 weightval1)
         ... )
       ( (ph3 ph4 ...)
         (weightname0 weightval0)
         (weightname1 weightval1)
         ... )
       ...
     )

These weights will typically be automatically generated.

Units INDEXFILE [PHONE_SET_NAME] UNITS

Compile given UNITS (in specified phoneme set) into INDEXFILE.

Use NAME

Use (and load if necessary the indexes of) udb NAME.

Keep

Save current database in list of databases. Create new (unset) database as current.

New NAME

Create a new database structure called 'NAME'.

Stats Unitnames

Return a list of unit names in current database.

Stats Occurences

Return a list of unit names plus number of occurrences.

Stats Units

Return total number of units.

Stats SampleRate

Return sample rate of waveforms.

Prune PList

If Plist is an atom, it is evaluated before use. It's value is a list of the form

     (phone_name prune_entry0 prune_entry1 ...)

where prune_entryN are database entries to be pruned from the index.

Display NAME

Display information about named database.

(define FUNCNAME (ARG0 ARG1 ...) EXPR0 EXPR1 ...)

Define function FUNCNAME. Note that functions are dynamically scoped.

(defvar VARNAME EXPR [DOCUMENTATION_STRING])

Set VARNAME (an atom) to the result of evaluating EXPR, if VARNAME is not already set. If DOCUMENTATION_STRING is set, it is stored and made available later for various help functions. Note even if VARNAME is already set, the documentation string is still set.

(Display COMMAND [OPTIONS])

COMMAND may be one of

Method: Type of Display, currently xmg or xwaves.
Open: Start display program. The third argument specifies what type of display (xmg, xwaves). If no third argument is present, the display type set by Method is used.
Close: Shut down display method.
Command: Send all following information to the display program.
UTT: Display the utterance (wave form, labelling etc). If any information follows, this is taken to be instructions on what should be displayed. The options can be any of wave, word, segment, intone, unit or f0. If options is NIL, or not specified, all output forms are generated.

(equal OBJ1 OBJ2)

Return 't' if OBJ1 and OBJ2 are equal, nil otherwise.

(error ERRMSG)

Print ERRMSG and cause an error.

(eval SEXPR)

Evaluate SEXPR and return the value.

(Feats_Out UTT STREAMNAME FEATURE-NAMES [OUTFILE])

Return a list of feature vectors for each cell in STREAMNAME of UTT. This is designed as a way to collect stats from existing utterances. The optional third argument specifies if the output is to be saved in a file. When OUTFILE is a file name, the features are saved (without parentheses) to that file. The features for each stream cell appear on different lines. If OUTFILE is "-", the output is printed to stdout. If OUTFILE is NIL, the output is returned as a list structure, which is the same as if no OUTFILE is specified. The list of features available can be obtained by calling Feats_Out with no arguments.

(Filter_Wave UTT FILTERNAME [optional arguments])

Run FILTERNAME of wave in UTT. If Filter_Wave is called with no arguments, a list of currently available filters is shown.

(flatten LIST)

Return a (new) flattened list from LIST, i.e. all sub-lists are removed, giving just a list of atoms.

(for INITIALIZATION ENDTEST LOOPINCREMENT EXPR0 ... EXPRN)

Iterative loop. Same as C's 'for'.

(free OBJ)

Do garbage collection on OBJ. There is no automatic garbage collection and this function is a poor substitute, but it does help out. If OBJ is an utterance, it will be garbage collected. Care should be taken in using this function.

(free LIST)

Free the con cells in LIST but not the members.

(free_val VARNAME)

Do garbage collect on the value of VARNAME, and set VARNAME's value to NIL. There is no automatic garbage collection and this function is a poor substitute, but it does help out. This function is safer than free, but it is still likely to cause segmentation faults if improperly used.

(function (ARG0 ARG1 ...) EXPR0 EXPR1 ...)

Return anonymous function. Mostly used in mapc and hooks.

(gc_info PARAM)

Set information for garbage collection type. Currently 'EnoughRope' is the only valid value. This causes aggressive garbage collection in utterances (at present specifically the arguments of utterances), which may cause your lisp code to fall over. If you don't understand this description, don't call this function.

(getenv VARNAME)

Return the value of VARNAME (an atom) in the UNIX environment.

(getpid)

Return the process ID. This is useful for ensuring your tmp files are unique.

(Help) or (Help FUNC_NAME)

Give general help or help for defined functions.

(HelpVar) or (Help VAR_NAME)

Give help on variables used within CHATR.

(History [Options] ...)

With no arguments displays the history list. Other arguments are

File FILENAME: Use FILENAME as history file.
Size [ NUM | unlimited ]: Set history size.
NUM: Display history of line number NUM. Negative numbers are treated as relative to current history, positive as absolute.

(if COND TRUE_EXPR [FALSE_EXPR])

If COND evaluates to non-nil, return value of TRUE_EXPR, otherwise return the value of FALSE_EXPR (or nil if there is no 'else' clause.

(Inspect UTT)

Use the graphical interface to inspect the contents of UTT.

(Int_Stats UTT OPTION0 OPTION1 ...)

Return statistics on intones in an utterance for results analysis. The OPTIONS describes what information is to be printed.

"E": Print all Events.
"C": Print all Connections.
"all": Print Connections and Events in order.
"Start": Print phrase start values.

(length LIST)

Return a number, the length of LIST (or 0 if it is an atom).

(let ((VAR0 VAL0) ...) EXPR0 ... EXPRN )

A command allowing introduction of local variables.

(Lexicon COMMAND ARG0 ARG1 ...)

Where COMMAND is one of

Compile INFILE OUTFILE: Compile lexicon in INFILE into OUTFILE.
Use FILE: Use compiled lexicon in file.
Add ENTRY0 ENTRY1 ...: Add lexical entries to addenda.
Lookup WORD: Lookup WORD in lexicon.
Fail [Error | xLTS | JLTS_LVO]: When a word is not found, either generate an error or use LTS letter-to-sound rules to guess. LTS is for English. JLTS implies a Japanese system. JLTS_LVO is for Japanese long vowels. KLTS is for Korean and GLTS for German.
Phone_Set NAME: Set phone set of current lexicon to NAME.
Select NAME: Select lexicon NAME or create new lexicon of that name if it does not already exist.
List: Return a list of the names of the currently defined lexicons.
Delete NAME: Delete lexicon NAME.

(Linear_Regression FILENAME)

Perform linear regression on the data in FILENAME. FILENAME should contain vectors, one per line, each line in parentheses. The first item in a vector is the item to be predicted, followed by the prediction factors (all floats). A correlation is printed, and a list returned consisting of the interceptor, plus weights for each factor.

(Linear_Regression_Detail FILENAME NAMES DETAIL)

DETAIL may be one of 3 values

0

Return a list of floats with the intercept first and the weights in the rest of the list.

1

Return a list containing an associated list with

     1.  an associated list for the weights
     2.  intercept
     3.  percentage variation
     4.  correlation
     5.  correlation of features to dependent variables

2

Return a list containing an associated list with 1-5 plus

     6.  weight * stdev for each feature
     7.  list of dropped features (weight = 0)
     8.  contribution of each feature to the model
     9.  stepwise contribution of features.

(Linear_Regression MODEL FILENAME [OUTFILE])

Use the given LR model to predict the values from the vectors in FILENAME. MODEL is a list of floats where the first item is the intercept and the remainder are the weights. FILENAME should consist of a list of vectors in the same format as input to Linear_Regression. Note these vectors still require that the first item be the value to be predicted. A list of predicted values is returned, or if OUTFILE specified they, are written to that file one per line (no parentheses). Correlation and statistics of mean error are printed.

(list OBJ1 OBJ2 ...)

Return a list constructed from the arguments.

(load FILENAME)

Load and evaluate commands in given file. Return the value of the last evaluated expression or nil if the file doesn't exist.

(load_kanji FILENAME)

Load kanji file and evaluate commands in given file.

(Load_Isard <index filename> <diphone filename>)

Load index and diphones for "Isard" synthesis method.

(load_library FILENAME)

Load and evaluate commands in FILENAME. The value of the variable 'load-path', a list of directories, is used to find the file. Each directory is checked for the given file in order. If FILENAME starts with a slash or tilde, or is not found in any library, the name is used directly. Returns the value of the last evaluated expression in the file, or nil if cannot find FILENAME.

(Load_Taylor)

Load index and perform initialization for 'Taylor' synthesis method. A number of lisp variables are used to describe the database

T_Index_Name

Name (and full path) of diphone index.

T_Dictionary_Name

Name (and full path) of diphone dictionary.

T_Vox_Path

Directory where waveform files are kept.

T_Pm_Path

Directory where pitchmark files are kept.

T_Sample_Rate

Sample rate of speech in Hz.

T_Diphone_Storage

Either 'GROUPED', implying all diphone waveforms and pitchmarks are compiled into a single dictionary, or 'SEPARATE', implying there is one waveform and pitchmark file per nonsense word.

T_Diphone_Type

Diphone waveforms can be coded in many ways, all affecting memory requirements. Note many of these options are now redundant or have not been fully incorporated into the system.

WAVEFORM: 16 bit pcm waveforms.
SHORTWAVEFORM: 16 bit pcm waveforms. (Not tested.)
FRAMES: Stored as separate frames in file. (Not tested.)
LPC: Stored as LPC coeffs. (Not tested.)
CODED_4: 16 ==> 4 bit compression.
CODED_5: 16 ==> 5 bit compression. (Not tested.)
CODED_6: 16 ==> 6 bit compression.
CODED_ALAW: 16 ==> 6 bit compression.
PITCH_LPC: Stored as pitch sync LPC coeffs. (Not tested.)
RES_LPC: Stored as pitch sync residule LPC. (Not tested.)

MAX_DIPHONES

Set the size of the internal cache of frequently used diphones. For a large machine, 100 - 500 is reasonable.

AVAILABLE_DIPHONES

Set the number of coded (and therefore available) diphones in RAM. (As opposed to MAX_DIPHONES, which sets the number of decoded diphones.)

(ls [OPTIONS] [FILENAME0 FILNAME1 ...])

Execute ls with arguments given.

(Make_PF INFILE OUTFILE)

Convert input file (similar to XLABEL format) to CHATR PhonoForm Utterance. Output is written to OUTFILE. If OUTFILE is "-", the output is written to standard output. If OUTFILE is NIL, OUTPUT is returned as a List structure. INFILE needs to be in a special format of time ordered events. It must contain phoneme, syllable, tone, word and break information. Scripts in the db creation suite make such files (and call this function).

(mapc FUNC ARG0_LIST ARG1_LIST ...)

Mapping function. Apply FUNC to an argument list built from the nth element of each argument list. Returns a list of results.

(Merge_Waves UTT1 UTT2)

Merge the waveform in UTT2 into that of UTT1.

(member ITEM LIST)

Return NIL if ITEM is not a member of LIST. If it is a member then the remainder of the list whose car is ITEM is returned. Uses equal rather than eq to determine membership.

(NN_Load NET)

NET is a list description of a net as generated by NN_Train (or other mechanism). Loads this description into the internal net structure and saves it in current_net. (what it does with the created net is likely to change.)

(NN_Test PAIRS_LIST/INFILE)

Test the currently loaded net with respect to the given pairs. Returns mean, rms and standard deviation of the difference between predicted and actual outputs.

(NN_Train PAIRS_LIST/INFILE OUTFILE ITERATIONS)

Train a neural network on given data. If PAIRS-LIST is atomic, it is treated as a file name. The file should contain i/o vectors, one per line (this is recommended for large training sets). If OUTFILE is nil then the net is returned as an s-expression. If OUTFILE is non-nil, the net is written (as an s-expression) to that file. The output format is suitable for the function NN_Load. The file will set a variable named from the basename OUTFILE with a list representation of the net. The variable nn_params is used to set auxiliary parameters.

(NN_use IVECTOR)

Find the output vector from this (atomic) IVECTOR using the currently loaded NET.

(not OBJ)

If OBJ is nil return 't', otherwise return nil.

(nth NTH LIST)

Return the NTH car from LIST, counting from 0. (This may be different from what you are used to.)

(NuuDB <command> <options>)

Set, compile, and generally manage NuuTALK databases. This is still pretty experimental and not a high priority to complete. See lib/data/nuutalk_*.ch for actual usage. 'command' may be one of

DumpBinIndex <fname>
LoadBinIndex <fname>
Select [name]: Select a database.
Keep: Store current database ready for selection.
Name <name>: Name current database.

(on_error ERROR_CASE EXPR)

Evaluate EXPR as usual, but if EXPR causes an error, catch it and evaluate ERROR_CASE then continue.

(or DISJ0 ...)

Evaluate conditions until one evaluates to non-nil. Return that or nil if no disjunct is non-nil.

(Output_Filename FILENAME_ROOT)

FILENAME_ROOT is used as the default root filename for 'Save' function calls. It will be appended by .wav .seg etc as appropriate.

(Parameter PARAMETER VALUE)

Set various synth parameters. With no arguments it returns the current configuration. PARAMETER may be one of

Synth_Method

Synthesis method to use. VALUE may be one of

UDB: Use unit database concatenation.
FORMANT_SYN: Use formant synthesis. (OPTIONAL)
ISARD: Use LPC diphone synthesis. (OPTIONAL)
TAYLOR: Use CSTR diphone synthesis. (OPTIONAL)
NUUTALK: Use (Japanese) NUUTALK cepstrum concatenation. (OPTIONAL)
NUUDB: Use (Japanese) NUUTALK cepstrum concatenation with original CHATR high level streams. (OPTIONAL)
NONE: Don't do any waveform synthesis.

Streams LIST

List of stream names to be used in utterances.

Int_Method

Intonation method to use. Resets Pitch_Stretch to 1.0. VALUE may be one of

CSTR: Same as RFC.
RFC
Tilt
JToBI: Japanese ToBI
Beckman: An alias for JToBI.
Fujisaki: Only for Japanese. Should be generic.
ToBI: Should be shared with Beckman.

Duration_Method

Duration method. Resets Duration_Stretch to 1.0. VALUE may be one of

KLATT_DUR
KAIKI_DUR: For Japanese.
NNet_dur: Multilangauge (Campbell-based) neural net system.
JVS_DUR: Jan van Santen POS module for f2b.
LR_DUR: Linear Regression method.
AVERAGE: Average phone durations.
AVERAGE+: Average phone duration with some simple alterations over syllables available for stressed/accented and phrase final.

Duration_Stretch x.x

Change overall duration by factor x.x.

Pitch_Stretch x.x

Change overall pitch by factor x.x.

Concat_Method

Unit concatenation method (where appropriate). VALUE may be one of

PSOLA: Helene's PSOLA.
PS_PSOLA: AWB's PSOLA.
NUUCEP: NUUTALK cepstrum re-synthesis.
CEPLMA: Cepstrum re-synthesis (same as NUUCEP).
DUMB: Direct concatenation.
DUMB+: Concatenation at minimum crossing distortion.
NONE: Don't concatenate anything, generate an empty wave. Useful in tests--saves reading any waveform files.

(Phoneme COMMAND ARG0 ARG1 ...)

With no argument prints a summary of the currently loaded phoneme sets. COMMAND may be one of

def PHONESET_NAME DEFINITIONS: Define phoneme set.
Map PHONESET_NAME1 PHONEME_SET_NAME2 MAPPING_LIST: Table showing how symbols in one phoneme set convert to symbols in the other.
Input_set PHONEME_SET_NAME): Set input to use phoneme set tagged by PHONEME_SET_NAME.
Internal_set PHONEME_SET_NAME): Set internal phoneme set to PHONEME_SET_NAME.
List PHONESET_NAME: Return a list of all phonemes in set.

(pprint_width NUM)

Set the pretty print width to NUM columns.

(predict_events UTT_SYL UTT_RFC)

Add E events to utt_syl based on the RFC elements in utt_rfc. This is a rather specific function used in Tilt labelling.

(print OBJ [FILENAME])

Print (and return) OBJ. If the optional third argument is given the result is printed to that file (and also returned). The function 'pprint_width' may be used to set the width used in formatting output.

(probe_file FILENAME)

Return the full path name of FILENAME if it can be found via the load-path, or nil if it cannot be found.

(prog EXPR0 EXPR1 ... EXPRN)

Evaluate EXPR0, ..., EXPRN and return value of EXPRN.

(provide PACKAGE-NAME)

State that PACKAGE-NAME is provided. (See require.)

(pwd)

Print current directory.

(quit)

Exit the CHATR system. Note that even when running in 'interactive' mode, and even if the variable chatr_confirm_exit is set to non-nil, this function still exists unconditionally. The function is designed to be used within CHATR programs rather than interactively.

(quote OBJ)

Return OBJ (unevaluated).

(Regain_Wave UTT FACTOR)

Normalise wave in UTT to maximum amplitude, then multiply by FACTOR. Default is 1.0 where 0.0 < FACTOR < 1.0.

(require PACKAGE-NAME)

If PACKAGE-NAME is not already provided (see provide), try to load a file (via load_library) whose name is made from 'PACKAGE_NAME' appended with '.ch'.

(Resamp_Wave UTT SAMP_RATE)

Resample the wave in UTT to SAMPLE_RATE, if not already in that sample rate.

(reverse LIST>)

Return the (destructively) reversed LIST.

(Sampling_Rate RATE)

Positive integer in Hz. Only used when no other more local specification of sample rate is available. Mainly used in reading ESPS headed files, and sometimes in reading or writing raw wave files.

(Save TYPE UTT [FILENAME])

Save utterance in the form defined by TYPE, which may be

Wave: Save waveform.
Pars: Save formant parameter file.
Segments: Save segment list.
Cep: Cepstrum parameters (in cep_file_type format).
F0: Fundamental frequency contour (xwaves label format).
XUnits: Select units and place where they joined in the waveform file (may not be exact) -- xwaves label format. This only makes sense if the units have been joined using simple dumb joins.
XSegs: (Target) Segments in xwaves labels format.
XWords: Words in xwaves format.
XTones: Intone info. (Note they might not be aligned to where you'd like them to be.)
XBreaks: Break index aligned to after each word.
F0: Fundamental frequency contour. If Display Method is XMG, this is in xmg format. Otherwise it is as an xwaves label file, not an Xwaves F0 file.
Intonation: Save a record of the intonation (not an utterance).
Events: Save the intonation events (not an utterance).
Connections: Save the intonation connections (not an utterance).
UnitLabels: Description of units selected.
UnitLabels+: Description of units selected.
NUSCand: Structure definitions for udb_nus strategy.
Romaji: Romaji Stream format.

If FILENAME is "-" then output is to stdout, except for waves or cepstrums. Because the authors of the system are incapable of remembering the order of the last two arguments, you can in fact specify them in any order. Thank us nicely, now.

(Say UTT)

Play waveform of UTT based on Audio Device settings.

(set VARNAME EXPR)

Set VARNAME (an atom) to the result of evaluating EXPR. NOTE this is really 'setq' in Lisp or 'set!' in Scheme, as it does not evaluate VARNAME.

(set! VARNAME EXPR)

Set VARNAME (an atom) to the result of evaluating EXPR.

(setenv VARNAME VALUE)

Set the UNIX environment variable VARNAME (an atom) to VALUE. (An atom typically a string.)

(Stats TYPE STATS)

Define statistics for a synthesis module. The stats can be of different types, where TYPE is one of

Duration: Phoneme duration statistics.
Intonation: Intonations statistics.
Pause: Pause statistics.
Klatt_dur: Durations specific for the Klatt formant synthesizer.
Pitch_range: Pitch range information.

(Stdin_To_File FILENAME EOF_MARK)

Read stdin until EOF_MARK found. Save the result in FILENAME. The function is designed for use in server mode, when text (including Kana/Kanji) may be streamed down a socket. CHATR has to capture that text up to the EOF_MARK and then continue with normal command reading mode.

(strcat ATOM1 ATOM2 ...)

Return an atom with a name made from the concatenation of the names of the atoms give as arguments.

(Synth UTT)

Synthesize UTT based on various parameters. This is the function that does all the work! The utterance object is destructively updated but also returned. This function will run other functions on the utterance as specified in the variable synth_hook, after waveform synthesis.

(system COMMAND)

Execute given Unix COMMAND via the bourne shell.

(train_input UTT_SYL UTT_RFC)

Build a (tilt) training file from a syllable utterance and an rfc utterance.

(Textfile FILENAME LANG)

A rather crude text to speech system. Will form utterances for all sentences in FILENAME. This is not really a user level function, as it requires various settings before it is useful. See the function 'tts'. LANG specifies the language to use or a list of escape commands and their translations to CHATR lisp functions. LANG may be English, Japanese, German or Korean. If LANG is not an atom, it may list escape commands. Each escape command is defined like a Lisp function, with a name, argument list, and a body. The next 'n' tokens are read as arguments to the function, where 'n' is the number of arguments. An example set of escape commands are

     (
       (@gsw () (speaker_eval a))
       (@duration (a)
	          (eval (list 'Parameter 'Duration_Stretch a)))
       (@pitch (a)
	          (eval (list 'Parameter 'Pitch_Stretch a)))
       (@slow () (Parameter Duration_Stretch 1.6))
       (@fast () (Parameter Duration_Stretch 0.6))
     )

(Udb_Best_N UTTERANCE N)

Return information about the best N paths found in unit selection.

(Udb_Clean_Training_Tables)

Clean up the tables created by Udb_Make_Training_Tables.

(Udb_Make_Training_Tables PHN_LIST)

Prepare the distance matrices and unit list for regression training with a given list of units. Calculate the acoustic distances for all unit pairs. Processes the currently loaded database and the currently defined cep_dist.

(Udb_Prune_On_Regression PHN_LIST)

Prune specified units using the acoustic distance measure and current distance weights.

(Udb_Stats UTTERANCE)

Return information about the unit selection within this utterance.

(Udb_Train_Context_Stats PHN_LIST [REG_ERROR])

Train the discrete tables of distance functions for the specified set of phonemes (in PHN_LIST), based on discrete variables. If REG_ERROR is specified then the discrete tables will be trained on the error in the current regression model, i.e. with the current weights.

(Udb_Train_Regression_Models PHN_LIST)

Determine weights for distances for the phone list.

(uload UTT EXPR)

Load the given utterance with the given Lisp description. This function is pretty new and may change. The description is as in the output of 'uprint'.

(uprint UTT [STREAM-LIST])

Return parts of the utterance object as s-expressions. Facilitates the print-out of the internal structure of an utterance object. With no options it prints out the input structure it is generating from. STREAM-LIST is an atom or a list of atoms identifying which 'streams' are to be output.

(Utterance TYPE FORM)

Create an utterance of TYPE from FORM. The result may be used for many other functions. Returns an object of type utterance. Note this function doesn't do much work. (Synth is the function that does all the work.) If utt_hook is set to a function or list of functions, these are applied to the utterance before being returned. See the User Manual for more details of the formats listed below. TYPE may be

HLP: High Level Prosody. Utterance is subjected to HLP rules.
PhonoWord: Phrasing, PitchRange and Intonation features may be specified.
Word: Simple word form.
Wave: A waveform.
Segment: Segment stream (as dumped by function Save Segments).
Unit: Naive concatenation of speech file units. Takes three forms of input.
SegF0: Segment level specification with F0 contour.
Syllable: Syllable level input. Many types, depending on intonation.
Text: Plain English text as a string.
Romaji: Kunreishiki Romaji with numbered break levels and accent marking. Similar to output from KDD program.

(get_segs UTT)

Return list of segments.

(stream_cell_count UTT [STREAM-LIST])

Add a Lisp function stream_cell_count which returns the number of cells in a named stream.

(udb_selection_Errors UTT)

Return a list of all segments with name, F0 error and duration error.

(Verbosity [Warning | Debug] [ON | OFF])

Set the verbosity level for warnings or debug information. 'ON' means the relevent messages are displayed. `OFF' means they aren't. Default is both OFF.

(Wave_Filetype FILETYPE)

The default method for saving waveforms. FILETYPE may be

ULAW: Raw 8KHz, ulaw, unheadered.
RAW: Raw **KHz shorts, in native byte order.
NIST: NIST Sphere headered. Supported by xwaves and naplay.
XMG: XMG format.

Go to the first, previous, next, last section, table of contents.