Skip to content

Commit b14378c

Browse files
JRMeyerJRMeyer
authored
Merge pull request #18 from Oct4Pie/main
Add Persian Model
2 parents 428bab2 + 90f5bdd commit b14378c

File tree

3 files changed

+357
-0
lines changed

3 files changed

+357
-0
lines changed

‎persian/oct4pie/v0.1.0/LICENSE

Lines changed: 165 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,165 @@
1+
GNU LESSER GENERAL PUBLIC LICENSE
2+
Version 3, 29 June 2007
3+
4+
Copyright (C) 2007 Free Software Foundation, Inc. <http://fsf.org/>
5+
Everyone is permitted to copy and distribute verbatim copies
6+
of this license document, but changing it is not allowed.
7+
8+
9+
This version of the GNU Lesser General Public License incorporates
10+
the terms and conditions of version 3 of the GNU General Public
11+
License, supplemented by the additional permissions listed below.
12+
13+
0. Additional Definitions.
14+
15+
As used herein, "this License" refers to version 3 of the GNU Lesser
16+
General Public License, and the "GNU GPL" refers to version 3 of the GNU
17+
General Public License.
18+
19+
"The Library" refers to a covered work governed by this License,
20+
other than an Application or a Combined Work as defined below.
21+
22+
An "Application" is any work that makes use of an interface provided
23+
by the Library, but which is not otherwise based on the Library.
24+
Defining a subclass of a class defined by the Library is deemed a mode
25+
of using an interface provided by the Library.
26+
27+
A "Combined Work" is a work produced by combining or linking an
28+
Application with the Library. The particular version of the Library
29+
with which the Combined Work was made is also called the "Linked
30+
Version".
31+
32+
The "Minimal Corresponding Source" for a Combined Work means the
33+
Corresponding Source for the Combined Work, excluding any source code
34+
for portions of the Combined Work that, considered in isolation, are
35+
based on the Application, and not on the Linked Version.
36+
37+
The "Corresponding Application Code" for a Combined Work means the
38+
object code and/or source code for the Application, including any data
39+
and utility programs needed for reproducing the Combined Work from the
40+
Application, but excluding the System Libraries of the Combined Work.
41+
42+
1. Exception to Section 3 of the GNU GPL.
43+
44+
You may convey a covered work under sections 3 and 4 of this License
45+
without being bound by section 3 of the GNU GPL.
46+
47+
2. Conveying Modified Versions.
48+
49+
If you modify a copy of the Library, and, in your modifications, a
50+
facility refers to a function or data to be supplied by an Application
51+
that uses the facility (other than as an argument passed when the
52+
facility is invoked), then you may convey a copy of the modified
53+
version:
54+
55+
a) under this License, provided that you make a good faith effort to
56+
ensure that, in the event an Application does not supply the
57+
function or data, the facility still operates, and performs
58+
whatever part of its purpose remains meaningful, or
59+
60+
b) under the GNU GPL, with none of the additional permissions of
61+
this License applicable to that copy.
62+
63+
3. Object Code Incorporating Material from Library Header Files.
64+
65+
The object code form of an Application may incorporate material from
66+
a header file that is part of the Library. You may convey such object
67+
code under terms of your choice, provided that, if the incorporated
68+
material is not limited to numerical parameters, data structure
69+
layouts and accessors, or small macros, inline functions and templates
70+
(ten or fewer lines in length), you do both of the following:
71+
72+
a) Give prominent notice with each copy of the object code that the
73+
Library is used in it and that the Library and its use are
74+
covered by this License.
75+
76+
b) Accompany the object code with a copy of the GNU GPL and this license
77+
document.
78+
79+
4. Combined Works.
80+
81+
You may convey a Combined Work under terms of your choice that,
82+
taken together, effectively do not restrict modification of the
83+
portions of the Library contained in the Combined Work and reverse
84+
engineering for debugging such modifications, if you also do each of
85+
the following:
86+
87+
a) Give prominent notice with each copy of the Combined Work that
88+
the Library is used in it and that the Library and its use are
89+
covered by this License.
90+
91+
b) Accompany the Combined Work with a copy of the GNU GPL and this license
92+
document.
93+
94+
c) For a Combined Work that displays copyright notices during
95+
execution, include the copyright notice for the Library among
96+
these notices, as well as a reference directing the user to the
97+
copies of the GNU GPL and this license document.
98+
99+
d) Do one of the following:
100+
101+
0) Convey the Minimal Corresponding Source under the terms of this
102+
License, and the Corresponding Application Code in a form
103+
suitable for, and under terms that permit, the user to
104+
recombine or relink the Application with a modified version of
105+
the Linked Version to produce a modified Combined Work, in the
106+
manner specified by section 6 of the GNU GPL for conveying
107+
Corresponding Source.
108+
109+
1) Use a suitable shared library mechanism for linking with the
110+
Library. A suitable mechanism is one that (a) uses at run time
111+
a copy of the Library already present on the user's computer
112+
system, and (b) will operate properly with a modified version
113+
of the Library that is interface-compatible with the Linked
114+
Version.
115+
116+
e) Provide Installation Information, but only if you would otherwise
117+
be required to provide such information under section 6 of the
118+
GNU GPL, and only to the extent that such information is
119+
necessary to install and execute a modified version of the
120+
Combined Work produced by recombining or relinking the
121+
Application with a modified version of the Linked Version. (If
122+
you use option 4d0, the Installation Information must accompany
123+
the Minimal Corresponding Source and Corresponding Application
124+
Code. If you use option 4d1, you must provide the Installation
125+
Information in the manner specified by section 6 of the GNU GPL
126+
for conveying Corresponding Source.)
127+
128+
5. Combined Libraries.
129+
130+
You may place library facilities that are a work based on the
131+
Library side by side in a single library together with other library
132+
facilities that are not Applications and are not covered by this
133+
License, and convey such a combined library under terms of your
134+
choice, if you do both of the following:
135+
136+
a) Accompany the combined library with a copy of the same work based
137+
on the Library, uncombined with any other library facilities,
138+
conveyed under the terms of this License.
139+
140+
b) Give prominent notice with the combined library that part of it
141+
is a work based on the Library, and explaining where to find the
142+
accompanying uncombined form of the same work.
143+
144+
6. Revised Versions of the GNU Lesser General Public License.
145+
146+
The Free Software Foundation may publish revised and/or new versions
147+
of the GNU Lesser General Public License from time to time. Such new
148+
versions will be similar in spirit to the present version, but may
149+
differ in detail to address new problems or concerns.
150+
151+
Each version is given a distinguishing version number. If the
152+
Library as you received it specifies that a certain numbered version
153+
of the GNU Lesser General Public License "or any later version"
154+
applies to it, you have the option of following the terms and
155+
conditions either of that published version or of any later version
156+
published by the Free Software Foundation. If the Library as you
157+
received it does not specify a version number of the GNU Lesser
158+
General Public License, you may choose any version of the GNU Lesser
159+
General Public License ever published by the Free Software Foundation.
160+
161+
If the Library as you received it specifies that a proxy can decide
162+
whether future versions of the GNU Lesser General Public License shall
163+
apply, that proxy's public statement of acceptance of any version is
164+
permanent authorization for you to choose that version for the
165+
Library.

‎persian/oct4pie/v0.1.0/MODEL_CARD.md

Lines changed: 99 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,99 @@
1+
# Model card for Persian STT v0.1.0
2+
3+
Jump to section:
4+
5+
- [Model card for Persian STT v0.1.0](#model-card-for-persian-stt-v010)
6+
- [Model details](#model-details)
7+
- [Intended use](#intended-use)
8+
- [Performance Factors](#performance-factors)
9+
- [Metrics](#metrics)
10+
- [Transcription Accuracy](#transcription-accuracy)
11+
- [Real-Time Factor](#real-time-factor)
12+
- [Model Size](#model-size)
13+
- [Approaches to uncertainty and variability](#approaches-to-uncertainty-and-variability)
14+
- [Training data](#training-data)
15+
- [Evaluation data](#evaluation-data)
16+
- [Ethical considerations](#ethical-considerations)
17+
- [Demographic Bias](#demographic-bias)
18+
- [Surveillance](#surveillance)
19+
- [Caveats and recommendations](#caveats-and-recommendations)
20+
21+
## Model details
22+
23+
- Person or organization developing model: Maintained by [oct4pie](https://github.com/Oct4Pie).
24+
- Model language: Persian / Farsi / `fa`, `fa-IR`
25+
- Model date: June 21, 2022
26+
- Model type: `Speech-to-Text`
27+
- Model version: `v0.1.0`
28+
- Compatible with 🐸 STT version: `v1.3.0`
29+
- License: GNU Lesser General Public License v3.0
30+
- Citation details: `@techreport{persian-stt, author = {Mehdi Hajmollaahmad Naraghi}, title = {Persian STT v0.1.0}, institution = {Coqui}, address = {\url{https://coqui.ai/models}} year = {2022}, month = {June}, number = {STT-FA-0.1.0} }`
31+
- [persian-tts](https://github.com/Oct4Pie/persian-stt) GitHub Repo
32+
- Where to send questions or comments about the model: You can leave an issue on [`STT` issues](https://github.com/coqui-ai/STT/issues), open a new discussion on [`STT` discussions](https://github.com/coqui-ai/STT/discussions), or chat with us on [Gitter](https://gitter.im/coqui-ai/).
33+
34+
## Intended use
35+
36+
Speech-to-Text for the [Persian Language](https://en.wikipedia.org/wiki/Persian_language) on 16kHz, mono-channel audio.
37+
38+
## Performance Factors
39+
40+
Factors relevant to Speech-to-Text performance include but are not limited to speaker demographics, recording quality, and background noise. Read more about STT performance factors [here](https://stt.readthedocs.io/en/latest/DEPLOYMENT.html#how-will-a-model-perform-on-my-data).
41+
42+
## Metrics
43+
44+
STT models are usually evaluated in terms of their transcription accuracy, deployment Real-Time Factor, and model size on disk.
45+
46+
### Transcription Accuracy
47+
48+
Using the language model with settings `lm_alpha=0.36669178512950323` and `lm_beta=0.3457913671678824` (found via `lm_optimizer.py`):
49+
50+
- Common-Voice clean: WER: 10.81\%, CER: 2.506\%
51+
- More about the model at [persian-tts repo](https://github.com/Oct4Pie/persian-stt)
52+
53+
### Real-Time Factor
54+
55+
Real-Time Factor (RTF) is defined as `proccesing-time / length-of-audio`. The exact real-time factor of an STT model will depend on the hardware setup, so you may experience a different RTF.
56+
57+
Recorded average RTF on laptop CPU: `.65`
58+
59+
60+
### Model Size
61+
62+
For STT, you always must deploy an acoustic model, and it is often the case you also will want to deploy an application-specific language model.
63+
64+
| Model type | Filename | Size |
65+
| ----------------------- | ---------------------- | ------ |
66+
| Acoustic model (tflite) | `persian_stt.tflite` | 45.3M |
67+
| Acoustic model (pb) | `persian_stt.pb` | 181M |
68+
| Acoustic model (pbmm) | `persian_stt.pbmm` | 181M |
69+
| Language model | `kenlm-persian.scorer` | 1.63GB |
70+
71+
### Approaches to uncertainty and variability
72+
73+
Confidence scores and multiple paths from the decoding beam can be used to measure model uncertainty and provide multiple, variable transcripts for any processed audio.
74+
75+
## Training data
76+
77+
This model was trained on the following corpora: Common Voice 9.0 Persian (cleaned and with custom train/dev/test splits). In total approximately ~271 hours of data.
78+
79+
## Evaluation data
80+
81+
The validation ("dev") sets were cleaned and generated from Common Voice 9.0 Persian.
82+
83+
## Ethical considerations
84+
85+
Deploying a Speech-to-Text model into any production setting has ethical implications. You should consider these implications before use.
86+
87+
### Demographic Bias
88+
89+
You should assume every machine learning model has demographic bias unless proven otherwise. For STT models, it is often the case that transcription accuracy is better for men than it is for women. If you are using this model in production, you should acknowledge this as a potential issue.
90+
91+
### Surveillance
92+
93+
Speech-to-Text may be mis-used to invade the privacy of others by recording and mining information from private conversations. This kind of individual privacy is protected by law in may countries. You should not assume consent to record and analyze private speech.
94+
95+
## Caveats and recommendations
96+
97+
Machine learning models (like this STT model) perform best on data that is similar to the data on which they were trained. Read about what to expect from an STT model with regard to your data [here](https://stt.readthedocs.io/en/latest/DEPLOYMENT.html#how-will-a-model-perform-on-my-data).
98+
99+
In most applications, it is recommended that you [train your own language model](https://stt.readthedocs.io/en/latest/LANGUAGE_MODEL.html) to improve transcription accuracy on your speech data.

‎persian/oct4pie/v0.1.0/alphabet.txt

Lines changed: 93 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,93 @@
1+
# Each line in this file represents the Unicode codepoint (UTF-8 encoded)
2+
# associated with a numeric index.
3+
# A line that starts with # is a comment. You can escape it with \# if you wish
4+
# to use '#' in the Alphabet.
5+
6+
'
7+
،
8+
؛
9+
؟
10+
ء
11+
آ
12+
أ
13+
ؤ
14+
ئ
15+
ا
16+
ب
17+
ة
18+
ت
19+
ث
20+
ج
21+
ح
22+
خ
23+
د
24+
ذ
25+
ر
26+
ز
27+
س
28+
ش
29+
ص
30+
ض
31+
ط
32+
ظ
33+
ع
34+
غ
35+
ـ
36+
ف
37+
ق
38+
ك
39+
ل
40+
م
41+
ن
42+
ه
43+
و
44+
ى
45+
ي
46+
ً
47+
ٌ
48+
َ
49+
ُ
50+
ِ
51+
ّ
52+
ْ
53+
ٔ
54+
٬
55+
پ
56+
چ
57+
ژ
58+
ک
59+
گ
60+
ۀ
61+
ی
62+
ے
63+
–
64+
“
65+
”
66+
…
67+
ﮐ
68+
ﮔ
69+
ﯾ
70+
ﯿ
71+
ﺍ
72+
ﺎ
73+
ﺑ
74+
ﺒ
75+
ﺖ
76+
ﺘ
77+
ﺧ
78+
ﺩ
79+
ﺪ
80+
ﺭ
81+
ﺮ
82+
ﺴ
83+
ﺷ
84+
ﺸ
85+
ﻋ
86+
ﻌ
87+
ﻢ
88+
ﻤ
89+
ﻥ
90+
ﻧ
91+
ﻭ
92+
ﻮ
93+
# The last (non-comment) line needs to end with a newline.

0 commit comments

Comments
 (0)