Member Avatar for lich

hi,

im doing my final year project. im using some api's. i was reasearching on speaker dependent systems. but i wanted to know about speaker independent AP's

but after searching for several hours i realised there are some api's so that the user do not have to do those put the words to the application. so those are Sphinx .

those its there i want to know if anyone have already used it or something. please help me. if anyone use it. tell me is it really working and is it really efficient.

as a example if we say 'test' from user1 and another user2 says 'test' will it detect as same 'test'.

please help me.

  • 2 Contributors
  • 2 Replies
  • 100 Views
  • 3 Days Discussion Span
  • Latest Post Latest Post by lich

Dani AI

Generated

As observed, toolkits like CMU Sphinx are one way to get a speaker‑independent recognizer without forcing each user to enroll. As noted, Microsoft has Windows speech APIs as well. A concise framing: "speaker‑independent" means the acoustic model is trained on many different voices so it can recognize unseen speakers. That works well for many tasks, but accuracy depends on model quality, acoustic conditions, and the length/ambiguity of the spoken material (single short words are harder to disambiguate than phrases).

Practical options today include mature open toolkits and commercial cloud services. Open/research stacks let a final‑year project experiment with model training and adaptation (for example Kaldi and the lightweight Vosk runtime). Cloud offerings trade convenience and high accuracy for cost and privacy considerations (for example Google Speech-to-Text, , Amazon Transcribe). CMU Sphinx remains usable for small or embedded projects (CMUSphinx).

Why the simple "test" example may vary: recognition uses an acoustic model + pronunciation lexicon + language model. Two speakers saying "test" will often be recognized as the same token on a decent speaker‑independent model, but short tokens are sensitive to noise, microphone, accent, and sampling rate. Ways to improve results: use a constrained grammar or keyword spotting for small vocabularies, capture 16 kHz mono audio with noise reduction/VAD, evaluate on a multi‑speaker test set and measure WER, and consider speaker‑adaptation techniques (available in Kaldi) only if more accuracy is needed.

Suggested project roadmap: pick an engine and run a baseline with its prebuilt models; collect a small evaluation set across multiple speakers and conditions and compute WER; if baseline is insufficient, try grammar/keyword approaches or offline toolkits with adaptation (Kaldi/Vosk), or use a cloud service for higher out‑of‑the‑box accuracy.

Recommended Answers

All 2 Replies

Member Avatar for ithelp

Microsoft has got some application for detecting text , you may check whether those apis are publicly available.

Member Avatar for lich

yes they are.. which is called SAPI. but that is not voice independent. its voice dependent. so u cant use it without feeding the data before the use

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.