top of page
jetagent2-transparent.gif
Jetlink Logo (3).png

One Step Further from Jetlink: Text-to-Speech with JetlinkTTS

  • Aug 17
  • 5 min read

Sound is no longer a tool, it's an experience.


How does a customer feel when they speak with your company's voice assistant? Do they feel like they're connecting with a robot or a real person?


This question is no longer just an aesthetic preference — it's an operational decision that directly impacts customer satisfaction, brand perception, and conversion rates. And for years, the answer was: "Good sound quality is expensive and not customizable." Jetlink is changing that equation. JetlinkTTS-v1.0-2B is now available as open source on Hugging Face — with support for 30 languages, voice cloning, creative sound design, and 48kHz studio quality.


What is Jetlink TTS?


JetlinkTTS is an AI model that converts text to human speech. But the term "text-to-speech" doesn't fully describe what this model does.

Technically, it's a tokenizer-free diffusion autoregressive model built on the VoxCPM2 architecture. Its 2B parameter, trained with over 2 million hours of multilingual speech data and offering 48kHz output quality, provides an infrastructure designed for enterprise applications.


Four features that truly make the difference:


  • Voice Design — You can create an entirely new voice personality using natural language description, without even needing a reference voice recording. You type in "a middle-aged, reliable, and calm male voice" — and the model produces that voice.

  • Voice Cloning — You can clone a person's voice from a short reference audio recording. By adding style controls, you can manipulate the tone, speed, and expression.

  • Ultimate Cloning — When you provide a transcript in addition to the reference audio recording, the model produces an output that closely resembles the original sound.

  • Streaming — Audio generation for real-time applications can operate in streaming mode; audio generation can begin before the text is completed.


30 Languages, One Model


For businesses targeting a global customer base, language support is a crucial criterion in choosing a TTS. JetlinkTTS supports 30 languages under a single model: Turkish, English, German, French, Arabic, Spanish, Japanese, Korean, Russian, Hindi, and more.


This means you can address different markets with a single infrastructure, without having to manage separate models for each language. For companies building a multilingual customer service operation, this feature reduces operational costs and makes it easier to create a consistent voice identity.


Looking specifically at Turkish: Given the scarcity of models that produce sufficiently natural, emphatic, and contextually responsive sounds for Turkish TTS, JetlinkTTS's support for this language is a significant advantage in itself.


In which scenarios is it used?


Voice Assistant and Customer Service


IVR systems, virtual assistants, voice versions of self-service chatbots — the quality of the voice spoken in all of these affects whether a customer proceeds. A monotonous, robotic voice discourages a customer from continuing to listen. With JetlinkTTS, you can create a unique voice identity for your brand and design different voice tones for different channels. A more sober voice for technical support, a more energetic tone for marketing communications — all from the same model.


Call Center Automation


In outgoing call campaigns, appointment reminders, or informational messages, converting text-based scripts into voice recordings is a significant operational burden. JetlinkTTS completely automates this process; you can generate the voice in minutes when you write a new script. With the voice cloning feature, you can clone from a reference voice you've designated as your brand voice, if any, and maintain a consistent voice identity across all your communication materials.


Content Creation and Voice Acting


Podcasts, e-learning modules, corporate training videos, product demos — professional voiceovers for all of these are costly and time-consuming. JetlinkTTS lets you automate much of this process. With the Ultimate Cloning feature, you can clone a voice recording you've already made and consistently use that same voice for new content. Studio booking, reshoots, audio editing — most of these processes are eliminated.


Localization and Subtitle Dubbing


When adapting video content to different languages, hiring a separate voiceover agency for each language increases both the cost and time. JetlinkTTS's support for 30 languages and consistent audio quality makes localization workflows much more efficient. You can have the same voice clone speak in different languages — a powerful option, especially for companies that use well-known voices, such as brand spokespersons.


Research and Development


For researchers working on speech synthesis, emotion detection, and multilingual voice systems, JetlinkTTS offers a production-quality starting point. With LoRA and full fine-tuning support, you can create a custom voice model for your domain with just 5–10 minutes of voice data.


Technical Infrastructure: Strong but Accessible


Jetlink TTS utilizes a particularly prominent feature of the VoxCPM2 architecture in terms of audio quality: the tokenizer-free diffusion autoregressive approach. This architecture significantly overcomes the "robotic" issues experienced in traditional TTS models and lays the groundwork for context-aware, expressive audio reproduction.


Hardware requirements:

  • Minimum ~8 GB VRAM for development and testing.

  • For more stable local operation, 12–16 GB of VRAM is recommended.

  • Datacenter-class GPUs for production and high concurrency.


Real-time performance: The RTX 4090 has a real-time factor of approximately 0.30; with Nano-VLLM acceleration, this drops to 0.13. This means the model operates much faster than the sound it produces — a critical feature for low-latency applications.


Setup:

pip install voxcpm

Basic use:


from voxcpm import VoxCPM
import soundfile as sf
model = VoxCPM.from_pretrained("Jetlink/JetlinkTTS-v1.0-2B", load_denoiser=False)

wav = model.generate(
    text="Merhaba, Jetlink yapay zeka platformuna hoş geldiniz.",
    cfg_value=2.0,
    inference_timesteps=10,
)

sf.write("output.wav", wav, model.tts_model.sample_rate)

Released under the Apache 2.0 license — commercially available, modifiable, and on-premises.


Fine-Tuning: Adjust Your Model to Your Own Taste


Using Jetlink TTS as is is a strong starting point. But fine-tuning it with your own data can make the model much more powerful.


The model supports both LoRA fine-tuning and full fine-tuning . If you have your own voiceover data, customer service conversations, or audio recordings containing domain-specific terminology, you can achieve a significant improvement with just 5–10 minutes of data.


This creates value especially in the following scenarios: companies targeting a specific accent or speaking style, those wanting to protect their brand voice, and those using industry-specific technical terminology.


Why Open Source?


We are applying the same principle here as with JetlinkSTT: Technology that transforms customer interactions must be transparent, testable, and improvable.


It's easy to offer a TTS model as a closed API. But this means developers can't fine-tune the model, researchers can't benchmark it, and enterprise teams can't host it on-premise according to data privacy requirements.

Releasing JetlinkTTS as open source eliminates all of these restrictions. You have complete control over the model.


Connecting with JetVice: The Platform's Audio Layer


JetlinkTTS forms the text-to-speech infrastructure for JetVoice , Jetlink's voice AI product . JetVoice delivers real-time voice experiences, human-like interactions in customer service, and workflow automation.


JetlinkSTT (speech-to-text) and JetlinkTTS (text-to-speech) together create a complete voice loop: the client speaks, the system understands, responds, and transmits in a natural voice. JetVoice offers this loop as a ready-to-use product at enterprise scale. Those who want to integrate the model into their own infrastructure can access it directly via Hugging Face. For those seeking a ready-made solution, JetVoice is the starting point.


Access the Model


JetlinkTTS-v1.0-2B is now in open access on Hugging Face:



To take your voice customer experience to the next level, create your brand voice, or explore how JetVoice can be used in your organization, speak with our team.


 
 
 

Comments

bottom of page