The speech manager interface is no longer supported and a replacement is

(unfortunately: especially the grail demo was hiarious) still missing.
This commit is contained in:
Jack Jansen 2002-12-13 15:25:26 +00:00
parent 72f3b7a5de
commit 3d87e3cdf7
3 changed files with 0 additions and 1463 deletions

File diff suppressed because it is too large Load Diff

View File

@ -1,228 +0,0 @@
# Pass this program the Holy Grail script on stdin.
import sys
import string
import stdwin
from stdwinevents import *
try:
import macspeech
except ImportError:
macspeech = None
WINWIDTH = 1000
scrw, scrh = stdwin.getscrsize()
if WINWIDTH > 0.8*scrw:
WINWIDTH = int(0.8*scrw)
BLACK = stdwin.fetchcolor('black')
RED = stdwin.fetchcolor('red')
BLUE = stdwin.fetchcolor('blue')
done='done'
class MacSpeaker:
def __init__(self):
self.voices = []
self.nvoices = macspeech.CountVoices()
self.curvoice = 1
self.rate = 1.0
def _newvoice(self):
vd = macspeech.GetIndVoice(self.curvoice)
sc = vd.NewChannel()
self.curvoice = self.curvoice + 1
if self.curvoice > self.nvoices:
self.curvoice = 1
return sc
def newvoices(self, n):
self.voices = []
for i in range(n):
self.voices.append(self._newvoice())
if self.rate <> 1.0:
self.setrate(1.0)
def setrate(self, factor):
self.rate = self.rate*factor
for v in self.voices:
r = v.GetRate()
v.SetRate(r*factor)
def speak(self, i, text):
self.voices[i-1].SpeakText(text)
def busy(self):
return macspeech.Busy()
[NOTHING, NEWSCENE, ACT, TEXT, MORETEXT] = range(5)
def parseline(line):
stripline = string.strip(line)
if not stripline:
return NOTHING, ''
if stripline[:5] == 'Scene':
return NEWSCENE, stripline
if line[0] == '[':
return ACT, stripline
if line[0] == ' ' and ':' in line:
splitline = string.splitfields(stripline, ':')
stripline = string.joinfields(splitline[1:], ':')
return TEXT, (splitline[0], string.strip(stripline))
return MORETEXT, stripline
def readscript(file):
lines = file.readlines()
acts = []
actor_dict = {}
longest = 0
prev_act = 0
for i in range(len(lines)):
tp, data = parseline(lines[i])
if tp == NEWSCENE:
acts.append((actor_dict.keys(), lines[prev_act:i]))
prev_act = i
actor_dict = {}
elif tp == TEXT:
actor_dict[data[0]] = 1
lines[i] = tp, data
return acts[1:]
class Main:
def __init__(self):
if macspeech:
self.speaker = MacSpeaker()
else:
self.speaker = None
sys.stdin = open('SCRIPT', 'r')
self.acts = readscript(sys.stdin)
maxactor = 0
for actorlist, actdata in self.acts:
if len(actorlist) > maxactor:
maxactor = len(actorlist)
if not self.loadnextact():
print 'No acts to play!'
raise done
self.lh = stdwin.lineheight()
self.winheight = (maxactor+2)*self.lh
stdwin.setdefwinsize(WINWIDTH, self.winheight)
self.win = stdwin.open('The Play')
self.win.setdocsize(WINWIDTH, self.winheight)
self.win.change(((0,0),(WINWIDTH, self.winheight)))
self.menu = self.win.menucreate('Play')
self.menu.additem('Faster', '+')
self.menu.additem('Slower', '-')
self.menu.additem('Quit', 'Q')
self.speed = 4
def done(self):
del self.win
del self.menu
def loadnextact(self):
if not self.acts: return 0
actors, lines = self.acts[0]
del self.acts[0]
prevactor = 0
for i in range(len(lines)):
tp, data = lines[i]
if tp == NOTHING:
continue
elif tp in (NEWSCENE, ACT):
lines[i] = 0, data
elif tp == TEXT:
prevactor = actors.index(data[0])
lines[i] = prevactor+1, data[1]
else:
lines[i] = prevactor+1, data
self.lines = lines
self.actors = [''] + actors
self.actorlines = [''] * len(self.actors)
if self.speaker:
self.speaker.newvoices(len(self.actors)-1)
self.prevline = 0
self.actwidth = 0
for a in self.actors:
w = stdwin.textwidth(a)
if w > self.actwidth:
self.actwidth = w
return 1
def loadnextline(self):
if not self.lines: return 0
self.actorlines[self.prevline] = ''
top = self.lh*self.prevline
self.win.change(((0, top), (WINWIDTH, top+self.lh)))
line, data = self.lines[0]
del self.lines[0]
self.actorlines[line] = data
self.prevline = line
top = self.lh*self.prevline
self.win.change(((0, top), (WINWIDTH, top+self.lh)))
if line == 0:
self.win.settimer(5*self.speed)
else:
if self.speaker:
self.speaker.speak(line, data)
tv = 1
else:
nwords = len(string.split(data))
tv = self.speed*(nwords+1)
self.win.settimer(tv)
return 1
def timerevent(self):
if self.speaker and self.speaker.busy():
self.win.settimer(1)
return
while 1:
if self.loadnextline(): return
if not self.loadnextact():
stdwin.message('The END')
self.win.close()
raise done
self.win.change(((0,0), (WINWIDTH, self.winheight)))
def redraw(self, top, bottom, draw):
for i in range(len(self.actors)):
tpos = i*self.lh
bpos = (i+1)*self.lh-1
if tpos < bottom and bpos > top:
draw.setfgcolor(BLUE)
draw.text((0, tpos), self.actors[i])
if i == 0:
draw.setfgcolor(RED)
else:
draw.setfgcolor(BLACK)
draw.text((self.actwidth+5, tpos), self.actorlines[i])
def run(self):
self.win.settimer(10)
while 1:
ev, win, arg = stdwin.getevent()
if ev == WE_DRAW:
((left, top), (right, bot)) = arg
self.redraw(top, bot, self.win.begindrawing())
elif ev == WE_TIMER:
self.timerevent()
elif ev == WE_CLOSE:
self.win.close()
raise done
elif ev == WE_MENU and arg[0] == self.menu:
if arg[1] == 0:
if self.speed > 1:
self.speed = self.speed/2
if self.speaker:
self.speaker.setrate(1.4)
elif arg[1] == 1:
self.speed = self.speed * 2
if self.speaker:
self.speaker.setrate(0.7)
elif arg[1] == 2:
self.win.close()
raise done
if 1:
main = Main()
try:
main.run()
except done:
pass
del main

View File

@ -1,31 +0,0 @@
#
# Hum - The singing macintosh
#
import macspeech
import sys
import string
dict = { 'A':57, 'A#':58, 'B':59, 'C':60, 'C#':61, 'D':62, 'D#':63,
'E':64, 'F':65, 'F#':66, 'G':67, 'G#':68}
vd = macspeech.GetIndVoice(2)
vc = vd.NewChannel()
print 'Input strings of notes, as in A B C C# D'
while 1:
print 'S(tr)ing-',
str = sys.stdin.readline()
if not str:
break
str = string.split(str[:-1])
data = []
for s in str:
if not dict.has_key(s):
print 'No such note:', s
else:
data.append(dict[s])
print data
for d in data:
vc.SetPitch(float(d))
vc.SpeakText('la')
while macspeech.Busy():
pass