Scholarlink

<p>In this paper, we introduce an approach for Translation of Telugu-to-English Speech through the process of translating speech from one language into speech in another language. Telugu is the Indian regional language among 22 constitutional languages. This work demonstrates translation of speech with a cascaded pipeline and models. Our Hybrid DNN-HMM acoustic model implemented for building Telugu ASR in Kaldi achieved WER score of 0.7 Percent when the inverse_language_model weight is 17 on the test dataset of Telugu speech database. We deployed large-scale speech data of 2628 utterances in Telugu language among 14 Indian regional languages as for speech data translation from Telugu language to English language. We present our Speech-to-Speech translation results that combine highly accurate Telugu Speech-to-Text (ASR), robust Google Translate machine translation (MT) engine and Gemini Vertex AI model (Text-to-Speech Synthesis) on Google Cloud. A cascade model is built for the translation of multiple language pairs from Telugu to English. The translated English language text rendered the translated synthesized English audio compared to the ground truth of the original Telugu audio. In this paper, we propose the Gemini Vertex AI Model on Google Cloud, which addresses the issues while translating English text to speech in Text-to-Speech synthesis by a) training synthesis model in a straight method using machine translated output than alternate text, and b) Specifying distinguished spoken elements like frequency, tone of voice, and exact time frame as the input Vertex AI. As a result, the wave file is generated by TTS model with the mentioned features. Our experimental results in this work represented the significance of the size of the training data that we used to build the Telugu ASR. We carried out this process for the translation from Telugu to English speech. This work is based on the choice of Indian regional languages and the translation of Indic languages in a multilingual scenario. The results also point out that human intervention is necessary in building ASR in Telugu language and Machine Translation for converting from Telugu Text to English Text. The goal is to achieve a High Quality Speech-to-Speech translation from the TTS Synthesis output in English language for better communication.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Translation of Telugu to English speech

  • 期刊论文

  • 发表日期：

    2026年2月20日

查看全文
  • Rama Rajeswari Mulukutla,
  • Suryakanth V. Gangashetty

摘要

In this paper, we introduce an approach for Translation of Telugu-to-English Speech through the process of translating speech from one language into speech in another language. Telugu is the Indian regional language among 22 constitutional languages. This work demonstrates translation of speech with a cascaded pipeline and models. Our Hybrid DNN-HMM acoustic model implemented for building Telugu ASR in Kaldi achieved WER score of 0.7 Percent when the inverse_language_model weight is 17 on the test dataset of Telugu speech database. We deployed large-scale speech data of 2628 utterances in Telugu language among 14 Indian regional languages as for speech data translation from Telugu language to English language. We present our Speech-to-Speech translation results that combine highly accurate Telugu Speech-to-Text (ASR), robust Google Translate machine translation (MT) engine and Gemini Vertex AI model (Text-to-Speech Synthesis) on Google Cloud. A cascade model is built for the translation of multiple language pairs from Telugu to English. The translated English language text rendered the translated synthesized English audio compared to the ground truth of the original Telugu audio. In this paper, we propose the Gemini Vertex AI Model on Google Cloud, which addresses the issues while translating English text to speech in Text-to-Speech synthesis by a) training synthesis model in a straight method using machine translated output than alternate text, and b) Specifying distinguished spoken elements like frequency, tone of voice, and exact time frame as the input Vertex AI. As a result, the wave file is generated by TTS model with the mentioned features. Our experimental results in this work represented the significance of the size of the training data that we used to build the Telugu ASR. We carried out this process for the translation from Telugu to English speech. This work is based on the choice of Indian regional languages and the translation of Indic languages in a multilingual scenario. The results also point out that human intervention is necessary in building ASR in Telugu language and Machine Translation for converting from Telugu Text to English Text. The goal is to achieve a High Quality Speech-to-Speech translation from the TTS Synthesis output in English language for better communication.