File size: 15,767 Bytes
c3b3053
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6cbf79e
c3b3053
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
00c0467
c3b3053
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
---
license: apache-2.0
library_name: transformers
tags:
  - automatic-speech-recognition
  - speech-to-text
  - asr
  - multilingual
  - audio
  - transformers
  - vllm
  - qwen
pipeline_tag: automatic-speech-recognition
base_model: Qwen/Qwen3-ASR-1.7B
---

# JetlinkSTT-3-1.7B

This repository hosts an organization-managed copy of **JetlinkSTT-3-1.7B** for speech recognition and language identification workloads.

It is intended for teams that want to manage deployment, access, and internal distribution from their own namespace while preserving compatibility with the upstream model ecosystem.

## Model Summary

JetlinkSTT-3-1.7B is a multilingual automatic speech recognition model designed for:

- **speech-to-text (ASR)**
- **language identification (LID)**
- **offline inference**
- **streaming inference**
- **long-audio transcription**

This model is based on the Qwen3-ASR family and is suitable for enterprise-grade transcription and speech understanding workflows.

## Key Features

- **Automatic speech recognition** for multilingual audio
- **Language identification** support
- **Streaming and offline unified inference**
- Strong robustness under:
  - noisy environments
  - challenging text patterns
  - singing voice
  - songs with background music
- Suitable for both research and production-oriented speech pipelines

## Language Support

The upstream Qwen3-ASR family supports **52 languages and dialects**, including **30 languages** and **22 Chinese dialects**.

Core language coverage includes:

- Chinese
- English
- Cantonese
- Arabic
- German
- French
- Spanish
- Portuguese
- Indonesian
- Italian
- Korean
- Russian
- Thai
- Vietnamese
- Japanese
- Turkish
- Hindi
- Malay
- Dutch
- Swedish
- Danish
- Finnish
- Polish
- Czech
- Filipino
- Persian
- Greek
- Hungarian
- Macedonian
- Romanian

The upstream family also supports multiple Chinese dialects and regional speech variations.

## Intended Use

This model is suitable for:

- multilingual speech transcription
- call center transcription
- meeting transcription
- subtitle generation
- voice assistant backends
- speech analytics pipelines
- speech data preprocessing
- research and benchmarking

## Model Details

### Architecture

JetlinkSTT-3-1.7B is based on **Qwen/Qwen3-ASR-1.7B**, part of the **Qwen3-ASR** family.

According to the upstream model card, the Qwen3-ASR family:

- leverages large-scale speech training data
- builds on the audio understanding capability of **Qwen3-Omni**
- supports unified streaming and offline inference
- is accompanied by a broader inference toolkit with support for batch inference, asynchronous serving, streaming, and timestamp-related workflows

## Hardware Requirements

> This model is not intended to be described with a single universal minimum hardware requirement.

Actual requirements depend on:

- inference backend
- batch size
- audio duration
- concurrency
- streaming vs offline mode
- latency targets
- memory overhead from the serving framework

### Minimum System Requirements

> The upstream model card does not define a single universal minimum VRAM requirement. Actual memory usage depends on the inference backend, audio duration, batch size, concurrency, streaming mode, and runtime configuration.

Estimated memory guidance for JetlinkSTT-3-1.7B:

- **Theoretical model weight footprint (BF16 / FP16):** approximately **3.4 GB VRAM**
- **Estimated practical minimum for light GPU inference:** **8 GB VRAM**
- **Recommended for smoother development and local testing:** **12–16 GB VRAM**
- **Recommended for production or higher-concurrency deployments:** datacenter-class GPUs

> Note: the values above are practical estimates, not hard universal limits. Real memory consumption can increase depending on longer audio inputs, larger batch sizes, streaming usage, and backend overhead.
### Reference Hardware

For practical deployment planning:

- **Development / light testing:** high-memory single GPU environments may be sufficient for basic experiments, depending on runtime settings
- **Production-oriented serving:** modern datacenter GPUs are recommended
- **Higher concurrency / lower latency deployments:** vLLM-backed or similarly optimized serving setups are recommended

### Practical Recommendation

For most teams:

1. start with isolated benchmarking
2. measure latency with your real audio durations
3. validate concurrency needs before production rollout
4. use optimized serving stacks when streaming or higher throughput is required

## Software Requirements

Recommended environment:

- Python 3.12
- Linux
- CUDA-enabled GPU environment
- One of the following workflows:
  - Transformers-based usage
  - `qwen-asr`
  - `qwen-asr[vllm]`

Common dependencies may include:

- `torch`
- `transformers`
- `qwen-asr`
- `flash-attn` for improved speed and reduced memory usage where supported

## Quickstart

Recommended environment setup:

    conda create -n jetlinkstt python=3.12 -y
    conda activate jetlinkstt

Minimal installation:

    pip install -U qwen-asr

For faster inference and streaming support:

    pip install -U qwen-asr[vllm]

Optional acceleration:

    pip install -U flash-attn --no-build-isolation

If your machine has limited RAM and many CPU cores:

    MAX_JOBS=4 pip install -U flash-attn --no-build-isolation

## Example Usage

Basic usage should follow the upstream Qwen3-ASR inference flow, adapted to your deployment environment and model path.

Example model reference:

    Jetlink/JetlinkSTT-3-1.7B

## Serving Notes

This model is suitable for:

- batch transcription
- asynchronous serving
- streaming speech recognition
- enterprise speech pipelines

For production deployments, consider:

- request queueing
- audio chunking strategy
- streaming transport design
- observability and latency monitoring
- GPU memory profiling
- concurrency benchmarking

## Strengths

- multilingual ASR support
- language identification capability
- unified streaming and offline inference
- strong robustness in difficult acoustic environments
- support for long-audio transcription
- suitable for enterprise speech workloads

## Limitations

- infrastructure needs depend heavily on concurrency and audio duration
- production serving characteristics vary by backend
- streaming deployments require additional engineering around chunking and transport
- real-world latency depends on model backend, hardware, and workload pattern

## Out-of-Scope / Cautionary Use

Outputs should be reviewed before use in:

- legal transcription without verification
- medical transcription without human review
- safety-critical voice workflows
- automated decisions based solely on raw transcription
- highly sensitive compliance scenarios without validation

Human review and downstream verification are strongly recommended.

## License

This repository follows the same license as the upstream release.

- **License:** Apache-2.0

If you redistribute, fine-tune, quantize, or otherwise modify this model, make sure your usage remains compliant with the upstream license and attribution requirements.

## Attribution

Original upstream model:
- `Qwen/Qwen3-ASR-1.7B`

This repository is an organization-managed copy and is **not the original upstream source**.

## Citation

Please cite the original Qwen3-ASR release when using this model in research, evaluation, or production documentation.

## Disclaimer

This repository may include packaging, naming, or deployment-oriented changes for organizational use.

For official updates, benchmark details, and upstream release notes, refer to the original Qwen model card.

---

# JetlinkSTT-3-1.7B (Türkçe)

Bu depo, konuşmadan metne dönüştürme ve dil tanıma iş yükleri için **JetlinkSTT-3-1.7B** modelinin kurum tarafından yönetilen bir kopyasını barındırır.

Bu depo; modeli kendi namespace’i altında yönetmek, erişimi kontrol etmek ve dağıtımı kolaylaştırmak isteyen ekipler için hazırlanmıştır. Amaç, upstream model ekosistemiyle uyumluluğu koruyarak kurumsal kullanım sağlamaktır.

## Model Özeti

JetlinkSTT-3-1.7B, aşağıdaki kullanım alanlarına yönelik çok dilli bir otomatik konuşma tanıma modelidir:

- **speech-to-text (ASR)**
- **language identification (LID)**
- **offline inference**
- **streaming inference**
- **uzun ses kayıtlarının transkripsiyonu**

Bu model, Qwen3-ASR ailesi temel alınarak hazırlanmıştır ve kurumsal seviyede konuşma tanıma iş akışları için uygundur.

## Temel Özellikler

- Çok dilli **otomatik konuşma tanıma**
- **Dil tanıma** desteği
- **Streaming ve offline birleşik inference**
- Aşağıdaki koşullarda güçlü dayanıklılık:
  - gürültülü ortamlar
  - zor metin örüntüleri
  - şarkı söyleme sesi
  - arka plan müzikli şarkılar
- Hem araştırma hem de production odaklı konuşma iş akışları için uygundur

## Dil Desteği

Upstream Qwen3-ASR ailesi **52 dil ve lehçe** desteği sunar. Buna **30 dil** ve **22 Çince lehçe** dahildir.

Başlıca desteklenen diller:

- Çince
- İngilizce
- Kantonca
- Arapça
- Almanca
- Fransızca
- İspanyolca
- Portekizce
- Endonezce
- İtalyanca
- Korece
- Rusça
- Tayca
- Vietnamca
- Japonca
- Türkçe
- Hintçe
- Malayca
- Hollandaca
- İsveççe
- Danca
- Fince
- Lehçe
- Çekçe
- Filipince
- Farsça
- Yunanca
- Macarca
- Makedonca
- Romence

Upstream aile ayrıca çeşitli Çince lehçeleri ve bölgesel konuşma varyasyonlarını da destekler.

## Kullanım Amacı

Bu model aşağıdaki senaryolar için uygundur:

- çok dilli konuşma transkripsiyonu
- çağrı merkezi konuşmalarının yazıya dökülmesi
- toplantı transkripsiyonu
- altyazı üretimi
- sesli asistan altyapıları
- konuşma analitiği süreçleri
- konuşma verisi ön işleme
- araştırma ve benchmark çalışmaları

## Model Detayları

### Mimari

JetlinkSTT-3-1.7B, **Qwen/Qwen3-ASR-1.7B** tabanlıdır ve **Qwen3-ASR** ailesinin bir parçasını temel alır.

Upstream model kartına göre Qwen3-ASR ailesi:

- büyük ölçekli konuşma verileriyle eğitilmiştir
- **Qwen3-Omni** modelinin ses anlama kabiliyetinden yararlanır
- birleşik streaming ve offline inference destekler
- batch inference, asynchronous serving, streaming ve timestamp odaklı iş akışlarını destekleyen daha geniş bir inference toolkit ile birlikte gelir

### Minimum Sistem Gereksinimleri

> Upstream model kartı tek ve evrensel bir minimum VRAM gereksinimi belirtmez. Gerçek bellek kullanımı; inference backend’ine, ses uzunluğuna, batch size’a, concurrency seviyesine, streaming moduna ve runtime ayarlarına göre değişir.

JetlinkSTT-3-1.7B için tahmini bellek rehberi:

- **Teorik model ağırlık boyutu (BF16 / FP16):** yaklaşık **3.4 GB VRAM**
- **Hafif GPU inference için tahmini pratik minimum:** **8 GB VRAM**
- **Daha rahat geliştirme ve local test için önerilen:** **12–16 GB VRAM**
- **Production veya daha yüksek concurrency gereken dağıtımlar için önerilen:** datacenter sınıfı GPU’lar

> Not: yukarıdaki değerler kesin ve evrensel sınırlar değil, pratik tahminlerdir. Daha uzun ses girdileri, daha büyük batch size, streaming kullanımı ve backend kaynaklı ek yükler gerçek bellek tüketimini artırabilir.
## Donanım Gereksinimleri

> Bu model için tek ve evrensel bir minimum donanım gereksinimi vermek doğru değildir.

Gerçek ihtiyaçlar şunlara bağlıdır:

- kullanılan inference backend’i
- batch size
- ses uzunluğu
- concurrency
- streaming veya offline kullanım
- gecikme hedefleri
- serving framework kaynaklı bellek yükü

### Referans Donanım

Pratik dağıtım planlaması için:

- **Geliştirme / hafif test:** runtime ayarlarına bağlı olarak yüksek belleğe sahip tek GPU ortamları temel denemeler için yeterli olabilir
- **Production odaklı serving:** modern datacenter GPU’lar önerilir
- **Daha yüksek concurrency / daha düşük latency gereken dağıtımlar:** vLLM tabanlı veya benzeri optimize serving yapıları önerilir

### Pratik Öneri

Çoğu ekip için en mantıklı yaklaşım:

1. önce izole benchmark almak
2. gerçek ses uzunluklarıyla latency ölçmek
3. production öncesi concurrency ihtiyacını doğrulamak
4. streaming veya yüksek throughput gerekiyorsa optimize serving stack kullanmak

## Yazılım Gereksinimleri

Önerilen ortam:

- Python 3.12
- Linux
- CUDA destekli GPU ortamı
- Aşağıdaki iş akışlarından biri:
  - Transformers tabanlı kullanım
  - `qwen-asr`
  - `qwen-asr[vllm]`

Yaygın bağımlılıklar şunları içerebilir:

- `torch`
- `transformers`
- `qwen-asr`
- uygun olduğunda hız ve bellek optimizasyonu için `flash-attn`

## Hızlı Başlangıç

Önerilen ortam kurulumu:

    conda create -n jetlinkstt python=3.12 -y
    conda activate jetlinkstt

Minimum kurulum:

    pip install -U qwen-asr

Daha hızlı inference ve streaming desteği için:

    pip install -U qwen-asr[vllm]

İsteğe bağlı hızlandırma:

    pip install -U flash-attn --no-build-isolation

Makinede RAM düşük ama CPU core sayısı fazlaysa:

    MAX_JOBS=4 pip install -U flash-attn --no-build-isolation

## Örnek Kullanım

Temel kullanım, upstream Qwen3-ASR inference akışına göre ve kendi deployment ortamına uyarlanarak yapılmalıdır.

Örnek model referansı:

    Jetlink/JetlinkSTT-3-1.7B

## Serving Notları

Bu model aşağıdaki kullanım türleri için uygundur:

- batch transcription
- asynchronous serving
- streaming speech recognition
- kurumsal konuşma iş akışları

Production dağıtımlarında şunlar değerlendirilmelidir:

- request queueing
- audio chunking stratejisi
- streaming transport tasarımı
- gözlemlenebilirlik ve latency takibi
- GPU bellek profillemesi
- concurrency benchmark’ları

## Güçlü Yönler

- çok dilli ASR desteği
- dil tanıma kabiliyeti
- birleşik streaming ve offline inference
- zorlu akustik ortamlarda güçlü dayanıklılık
- uzun ses kayıtlarının transkripsiyonuna uygunluk
- kurumsal konuşma iş yükleri için uygun yapı

## Sınırlamalar

- altyapı ihtiyacı concurrency ve ses uzunluğuna göre ciddi şekilde değişir
- production serving karakteristiği kullanılan backend’e göre değişir
- streaming dağıtımlarında chunking ve taşıma katmanı için ek mühendislik gerekir
- gerçek dünya gecikmesi; backend, donanım ve iş yükü desenine bağlıdır

## Kapsam Dışı / Dikkat Gerektiren Kullanımlar

Model çıktıları şu senaryolarda insan kontrolü olmadan kullanılmamalıdır:

- doğrulamasız hukuki transkripsiyon
- insan incelemesi olmadan tıbbi transkripsiyon
- güvenlik kritik ses iş akışları
- yalnızca ham transkripsiyona dayalı otomatik kararlar
- doğrulama olmadan yüksek hassasiyetli uyumluluk senaryoları

İnsan incelemesi ve aşağı akış doğrulama mekanizmaları güçlü şekilde önerilir.

## Lisans

Bu depo, upstream sürümle aynı lisansı takip eder.

- **Lisans:** Apache-2.0

Modeli yeniden dağıtıyor, fine-tune ediyor, quantize ediyor veya başka şekilde değiştiriyorsan; kullanımının upstream lisans ve attribution gereklilikleriyle uyumlu olduğundan emin olmalısın.

## Atıf

Orijinal upstream model:
- `Qwen/Qwen3-ASR-1.7B`

Bu depo, kurum tarafından yönetilen bir kopyadır ve **orijinal upstream kaynak değildir**.

## Atıf / Citation

Bu modeli araştırma, değerlendirme veya production dokümantasyonunda kullanıyorsan, lütfen orijinal Qwen3-ASR sürümüne atıf yap.

## Feragatname

Bu depo, kurumsal kullanım amacıyla paketleme, isimlendirme veya dağıtım odaklı bazı değişiklikler içerebilir.

Resmi güncellemeler, benchmark detayları ve upstream sürüm notları için orijinal Qwen model kartına bakılmalıdır.