;============================================================================== ; VTX Speech player for the Vectrex (AY-3-8912 formant speech). AS09 syntax. ; Version 1.0, MIT (see LICENSE). Background: https://mikiex.github.io/vtx-speech/ ; ; Envelope shape 8 is the glottal buzz at the voice pitch; A and B play F1/F2 ; as harmonics of it; noise on B for consonants. Registers are set once per ; 50 Hz frame from word tracks written by export.py / the editor. ; ; Define before including: ; VS_RAM equ $CA80 ; 22 bytes ; VS_DUCK equ 4 ; channel C volume reduction while speaking (0 = none) ; ; include "vtx_speech.asm" ; include "speech_data.asm" ; ; Routines (all may clobber A, B, X; DP any on entry, preserved): ; Speech_Init once ; Speech_Say X = phrase label; replaces any playing phrase ; Speech_Busy Z = 1 when silent ; Speech_Stop silence now ; Speech_Service once per frame, after your sound code fills ; Vec_Music_Work and before Do_Sound ; Speech_Wait_Recal optional Wait_Recal replacement: counts late 50 Hz ; frames (VIA T2) so speech keeps real time ; ; While speaking the player owns R0-R3, R6, R8, R9, R11-R13 and the A/B bits ; of R7. C is yours, reduced by VS_DUCK. At a phrase end or Speech_Stop, A and ; B are disabled in R7 and set to volume 0. ; ; Cost (MAME): 642 cycles/frame mean while speaking, 812 p95, 1,180 max. ; Checked frame by frame in MAME against the reference player (3,733 frames). ;============================================================================== ;--- BIOS (own names, so this file never clashes with your equates) ---------- VS_BIOS_Wait_Recal equ $F192 VS_BIOS_Random equ $F517 ; needs DP = $C8 VS_VIA_T2_HIGH equ $D009 ; VIA timer 2, high byte VS_VIA_IFR equ $D00D ; VIA interrupt flags (bit 5: timer 2 done) ;--- The BIOS sound buffer: Vec_Music_Work holds R13 at $C83F up to R0 at $C84C. ; Do_Sound copies it to the chip, writing only registers that differ from ; the shadow copy at Vec_Snd_Shadow ($C800 = R0 ... $C80D = R13). VS_R0 equ $C84C VS_R1 equ $C84B VS_R2 equ $C84A VS_R3 equ $C849 VS_R6 equ $C846 VS_R7 equ $C845 VS_R8 equ $C844 VS_R9 equ $C843 VS_R10 equ $C842 VS_R11 equ $C841 VS_R12 equ $C840 VS_R13 equ $C83F VS_Shadow_R13 equ $C80D ;--- RAM (22 bytes from VS_RAM) --------------------------------------------- VS_Phrase equ VS_RAM+0 ; 2: the next phrase byte VS_Word equ VS_RAM+2 ; 2: the next byte of the word (0: between words) VS_Left equ VS_RAM+4 ; frames left in the current run VS_Head equ VS_RAM+5 ; the run's header byte (bits 5-4 kind, bit 6 noise) VS_Q1 equ VS_RAM+6 ; F1 in 16 Hz units VS_Q2 equ VS_RAM+7 ; F2 in 16 Hz units VS_Noise equ VS_RAM+8 ; index into VS_Noises VS_Clock equ VS_RAM+9 ; the voice 0..7 VS_Voiced equ VS_RAM+10 ; the last frame written was voiced VS_Ticks equ VS_RAM+11 ; 50 Hz frames to play at the next service VS_Late equ VS_RAM+12 ; lateness carried over, in 256-cycle units VS_Ep equ VS_RAM+13 ; envelope period of the voice VS_Mixer equ VS_RAM+14 ; the A and B mixer bits being built VS_Tmp equ VS_RAM+15 ; 2: scratch VS_Steps equ VS_RAM+17 ; frames still to step in this service VS_Return equ VS_RAM+18 ; 2: where a chosen alternative returns to (0: none) VS_State equ VS_RAM+20 ; 2: a phrase to start, or VS_Busy, or VS_Done ;--- Two marker bytes: VS_State points at one of them, or at a new phrase. VS_Busy: fcb $00 ; a phrase is playing VS_Done: fcb $FF ; silent ; Reciprocals for the harmonic lock: R[n] = round(16384 / n), n = 0..63. VS_Recip: fdb 0,16384,8192,5461,4096,3277,2731,2341,2048,1820,1638,1489,1365,1260,1170,1092 fdb 1024,964,910,862,819,780,745,712,683,655,630,607,585,565,546,529 fdb 512,496,482,468,455,443,431,420,410,400,390,381,372,364,356,349 fdb 341,334,328,321,315,309,303,298,293,287,282,278,273,269,264,260 ;============================================================================== ; Speech_Init: call once before anything else. ;============================================================================== Speech_Init: ldx #VS_Done stx >VS_State lda #1 sta >VS_Ticks clr >VS_Late clr >VS_Voiced rts ;============================================================================== ; Speech_Say: X = a phrase. It starts at the next Speech_Service. ;============================================================================== Speech_Say: stx >VS_State rts ;============================================================================== ; Speech_Busy: Z set when silent (nothing playing and nothing waiting). ;============================================================================== Speech_Busy: ldx >VS_State cmpx #VS_Done rts ;============================================================================== ; Speech_Stop: silence now. Channels A and B go off: mixer and volume. ;============================================================================== Speech_Stop: ldx #VS_Done stx >VS_State bra VS_Silence ; Silence A and B. Switching tone and noise off is not enough: a channel ; with both off still outputs its volume, and a volume that follows the ; envelope keeps playing the envelope (a buzz at the last pitch). So the ; volumes go to 0 too; your next sound sets its own. VS_Silence: lda >VS_R7 ora #$1B ; A and B: tone and noise off sta >VS_R7 clr >VS_R8 ; A and B: volume 0 clr >VS_R9 clr >VS_Voiced rts ;============================================================================== ; Speech_Wait_Recal: use instead of the BIOS Wait_Recal (optional). ; Wait_Recal waits for VIA timer 2, which it set for 30,000 cycles (one ; 50 Hz frame) at the start of the last frame. If the timer has already run ; out, the frame was late: the timer keeps counting down past zero, so its ; high byte, negated, is how late in 256-cycle units. Each 117 units ; (29,952 cycles) is one more 50 Hz frame for speech to step. Reading the ; high byte leaves the timer flag for Wait_Recal. ;============================================================================== Speech_Wait_Recal: lda >VS_VIA_IFR bita #$20 ; timer 2 done: the frame ran late beq VS_On_Time ldb >VS_VIA_T2_HIGH ; counting down past zero negb ; cycles late / 256 (rounded up) clra addb >VS_Late adca #0 VS_Tick_Loop: cmpd #117 ; 117 * 256 = 29,952 cycles: one more frame blo VS_Tick_Done subd #117 inc >VS_Ticks bra VS_Tick_Loop VS_Tick_Done: stb >VS_Late ; keep the remainder for next time VS_On_Time: jmp >VS_BIOS_Wait_Recal ;============================================================================== ; Speech_Service: once a frame. Steps the phrase by the frames that have ; passed (1, or more after a late frame with Speech_Wait_Recal) and writes ; the registers of the last one into the BIOS sound buffer. ;============================================================================== Speech_Service: ldx >VS_State cmpx #VS_Busy beq VS_Continue cmpx #VS_Done beq VS_Idle ; A new phrase (Speech_Say stored its address). Its first frame plays now. stx >VS_Phrase ldx #VS_Busy stx >VS_State clr >VS_Word clr >VS_Word+1 clr >VS_Return clr >VS_Return+1 clr >VS_Left clr >VS_Clock ; voice 0 until the phrase sets one lda #1 bra VS_Steps_Set VS_Continue: lda >VS_Ticks VS_Steps_Set: sta >VS_Steps VS_Step_Loop: bsr VS_Step ; carry: the phrase has ended bcs VS_Ended dec >VS_Steps bne VS_Step_Loop lda #1 sta >VS_Ticks jmp >VS_Write VS_Ended: ldx #VS_Done stx >VS_State bsr VS_Silence VS_Idle: lda #1 sta >VS_Ticks rts ;------------------------------------------------------------------------------ ; VS_Step: one 50 Hz frame. When the current run is used up, reads phrase ; tokens until the next run starts. Carry set at the end of the phrase. ; ; Phrase bytes: $00-$3F say word n $80-$87 voice 0..7 ; $88 a random voice $89 n a1..an one of n alternatives ; $C1-$FE silence, 1..62 frames $FF end ;------------------------------------------------------------------------------ VS_Step: tst >VS_Left lbne VS_Have ; the run goes on ldx >VS_Word bne VS_Header ; the next run of this word VS_Token: ldx >VS_Phrase lda ,x+ stx >VS_Phrase cmpa #$80 blo VS_Start_Word ; $00-$7F: a word cmpa #$FF beq VS_Seq_End cmpa #$C0 bhi VS_Pause ; $C1-$FE: silence cmpa #$89 bhs VS_Choice ; $89, $8A: a random choice cmpa #$88 bne VS_Set_Clock ; $80-$87: a voice lbsr VS_Random ; $88: a random voice VS_Set_Clock: anda #7 sta >VS_Clock bra VS_Token ; $FF: the end of an alternative (back to where it was chosen) or of the phrase. VS_Seq_End: ldx >VS_Return lbeq VS_End stx >VS_Phrase clr >VS_Return clr >VS_Return+1 bra VS_Token ; $89 n refs: say one of n alternatives, then go on after the references. ; ($8A n refs: go on with one of n phrases, no return.) Choice = (r * n) >> 8. VS_Choice: ldb ,x+ ; n pshs a,b stx >VS_Phrase ; the references (VS_Random changes X) lbsr VS_Random ldb 1,s mul ; A = (r * n) >> 8, below n lsla ; n <= 32: offset < 64 ldx >VS_Phrase ldd a,x std >VS_Tmp ; the chosen sequence puls a,b cmpa #$89 bne VS_Jump lslb abx ; after the n references stx >VS_Return VS_Jump: ldx >VS_Tmp stx >VS_Phrase bra VS_Token VS_Pause: suba #$C0 sta >VS_Left clr >VS_Head ; silent, no noise bra VS_Have VS_Start_Word: lsla ; word numbers < 64: offset < 128 (signed) ldx #VS_Words ldx a,x ; A run's header: bit 7 formant bytes follow (F1, then F2 unless kind 3), ; bit 6 a noise byte follows, bits 5-4 kind (0 silent, 1 voiced, 2 nasal, ; 3 F1 only), bits 3-0 frames - 1. $FF ends the word. VS_Header: lda ,x+ cmpa #$FF bne VS_Run clr >VS_Word ; the word's end clr >VS_Word+1 bra VS_Token VS_Run: sta >VS_Head bita #$80 ; formants follow (else the last run's stay) beq VS_No_Formants ldb ,x+ stb >VS_Q1 anda #$30 cmpa #$30 ; kind 3: F1 only beq VS_No_Formants ldb ,x+ stb >VS_Q2 VS_No_Formants: lda >VS_Head bita #$40 beq VS_No_Noise ldb ,x+ stb >VS_Noise VS_No_Noise: stx >VS_Word ; Frames at this voice: max(1, (n * S + 64) >> 7), S = 8 * (16 - voice). ; Low voices are slower, as on the arcade's speech chip; no run vanishes. anda #$0F inca ldx #VS_Clock_Scale ldb >VS_Clock ldb b,x mul addd #64 lslb rola bne VS_Length inca VS_Length: sta >VS_Left VS_Have: dec >VS_Left andcc #$FE rts VS_End: orcc #$01 rts ; A = a random byte (the BIOS Random needs DP = $C8). VS_Random: pshs dp lda #$C8 tfr a,dp jsr >VS_BIOS_Random VS_Random_Got: ; (the MAME check logs A here) puls dp,pc ;------------------------------------------------------------------------------ ; VS_Write: the current frame's registers into the BIOS sound buffer. ; voiced (kinds 1-3): A plays F1 on the envelope ; kind 1 (voiced): B plays F2 on the envelope ; kind 2 (nasal): B plays F2 at a steady, quiet volume 5 ; noise: B plays noise (period and volume from VS_Noises) ; the envelope: sawtooth (shape 8) at the voice's pitch ;------------------------------------------------------------------------------ VS_Write: ldx #VS_Clock_EP ldb >VS_Clock lda b,x sta >VS_Ep lda #$1B ; our mixer bits: A and B tone and noise off sta >VS_Mixer ldd #0 std >VS_R1 ; R1, R0 std >VS_R3 ; R3, R2 sta >VS_R8 sta >VS_R9 sta >VS_R6 lda >VS_Head anda #$30 beq VS_A_Done ; silent: A off lda >VS_Q1 ; voiced: F1 on A, on the envelope lbsr VS_Tone std >VS_R1 lda #$10 sta >VS_R8 lda #$1A ; tone A on sta >VS_Mixer VS_A_Done: lda >VS_Head bita #$40 bne VS_B_Noise anda #$30 beq VS_Mix cmpa #$30 beq VS_Mix ; F1 only: B off ldb #$10 ; voiced: F2 on the envelope cmpa #$10 beq VS_B_Volume ldb #5 ; nasal: F2 steady and quiet VS_B_Volume: stb >VS_R9 lda >VS_Q2 lbsr VS_Tone std >VS_R3 lda >VS_Mixer anda #$FD ; tone B on sta >VS_Mixer bra VS_Mix VS_B_Noise: ldb >VS_Noise lslb ldx #VS_Noises abx ldd ,x ; A = noise period, B = volume sta >VS_R6 stb >VS_R9 lda >VS_Mixer anda #$EF ; noise B on sta >VS_Mixer VS_Mix: lda >VS_R7 anda #$24 ; keep channel C's bits; bits 6-7 stay 0 ora >VS_Mixer ; (the Vectrex reads its buttons through port A) sta >VS_R7 lda >VS_Ep sta >VS_R11 clr >VS_R12 lda #8 ; envelope shape 8: repeating sawtooth sta >VS_R13 lda >VS_Head anda #$30 beq VS_Unvoiced tst >VS_Voiced bne VS_Duck ; A voice onset: writing R13 restarts the envelope, so the first pitch ; period starts loud. Do_Sound writes a register only when it differs ; from the shadow, so spoil the shadow's R13 to force the write. lda #$FF sta >VS_Shadow_R13 lda #1 sta >VS_Voiced bra VS_Duck VS_Unvoiced: clr >VS_Voiced VS_Duck: lda >VS_R10 ; your channel C, VS_DUCK steps down suba #VS_DUCK bcc VS_Duck_Set clra VS_Duck_Set: sta >VS_R10 rts ;------------------------------------------------------------------------------ ; VS_Tone: D = the tone period for the formant in A (16 Hz units) at the ; envelope period VS_Ep, moved to the nearest harmonic of the pitch, so the ; tone repeats in step with every pitch period (a clean buzz): ; n = (q * ep * 179 + 32768) >> 16, clamped to 1..63 (nearest harmonic) ; TP = (ep * R[n] + 512) >> 10 (= 16 * ep / n) ; No division: four 8x8 MULs and the reciprocal table. Changes A, B, X. ;------------------------------------------------------------------------------ VS_Tone: ldb >VS_Ep mul ; P = q * ep pshs a ; P high lda #179 mul ; P low * 179 sta >VS_Tmp ; its high byte puls b lda #179 mul ; P high * 179 addb >VS_Tmp adca #0 addb #128 ; + 32768 >> 8 adca #0 ; A = n bne VS_N_Low inca VS_N_Low: cmpa #63 bls VS_N_Ok lda #63 VS_N_Ok: lsla ldx #VS_Recip ldd a,x ; R[n] pshs b ; R low ldb >VS_Ep mul ; ep * R high std >VS_Tmp puls a ldb >VS_Ep mul ; ep * R low tfr a,b clra addd >VS_Tmp ; (ep * R + 512) >> 8 = ep*Rh + (ep*Rl >> 8) + 2 addd #2 lsra rorb lsra rorb ; >> 2 rts