# 12 Best MiniMax Speech Alternatives (2026)

> 12 verified MiniMax Speech alternatives in Text-to-speech APIs, led by Fish Audio. Compared on real production cost and per-use-case verdicts. Updated.

MiniMax Speech is a multilingual voice api from chinese ai lab minimax with strong voice cloning, 40+ languages, and emotion controls - sold pay-as-you-go or through cheap credit subscriptions starting at $5/month.. Teams that switch usually cite price at production volume, language coverage, self-hosting and license control. The alternatives below are ranked by published head-to-head verdicts, not sponsorship.

On our facts, MiniMax Speech is the 10th-cheapest fast-model rate of 10 and the 15th-cheapest flagship rate of 16 - the kind of gap teams cite when they go looking.

## Where to switch, by reason

| Switching because of | See |
| --- | --- |
| price at production volume | [Cartesia](https://www.versusref.com/tts/best/developers/) |
| language coverage | [Inworld TTS](https://www.versusref.com/tts/best/dubbing-localization/) |
| self-hosting and license control | [Voxtral TTS](https://www.versusref.com/tts/best/self-hosted/) |
| audiobooks | [Azure Speech](https://www.versusref.com/tts/best/audiobooks/) |
| content creators | [ElevenLabs](https://www.versusref.com/tts/best/content-creators/) |

## The MiniMax Speech alternatives, ranked

| # | Tool | Positioning | vs MiniMax Speech | Best if you need | Price | Rating |
| --- | --- | --- | --- | --- | --- | --- |
| 1 | [Fish Audio](https://www.versusref.com/tts/tools/fish-audio/) | Developer-first hosted API from the team behind the open-source fish-speech models; simple prepaid pay-as-you-go billing, 80+ language coverage, and a free fair-use model tier (s2.1-pro-free). Facts here cover the hosted API, not the OSS model. | vs MiniMax Speech: ~2× the languages, ~85% cheaper. | audiobooks, dubbing, self-hosted, voice agents | See pricing |  |
| 2 | [Amazon Polly](https://www.versusref.com/tts/tools/amazon-polly/) | Cloud-utility TTS at commodity prices | vs MiniMax Speech: ~95% cheaper, ~70% cheaper, no instant voice cloning. |  | See pricing |  |
| 3 | [Azure Speech](https://www.versusref.com/tts/tools/azure-speech/) | Enterprise hyperscaler TTS with custom-voice depth | vs MiniMax Speech: ~2.5× the languages, ~80% cheaper. |  | From $960/mo |  |
| 4 | [CAMB.AI](https://www.versusref.com/tts/tools/camb-ai/) | Localization-first TTS: the MARS 8 family (flash/pro/instruct variants) plus dubbing and translated-TTS pipelines, credit-based plans from $5/mo, aimed at media, sports, and content going multilingual. | vs MiniMax Speech: ~4× the languages. |  | From $5/mo |  |
| 5 | [Cartesia](https://www.versusref.com/tts/tools/cartesia/) | Lowest-latency TTS for real-time voice agents | vs MiniMax Speech: ~65% faster. |  | From $5/mo |  |
| 6 | [Chatterbox](https://www.versusref.com/tts/tools/chatterbox/) (OSS) | Production-minded open TTS from a commercial voice company - MIT license, built-in watermarking, and a fast Turbo variant, with Resemble's paid API as the scale-up path. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 7 | [ChatTTS](https://www.versusref.com/tts/tools/chattts/) (OSS) | Optimized for natural dialogue-style speech for LLM assistants; the licensing combination (AGPLv3+ code, CC BY-NC 4.0 weights, research/education only) rules out most commercial SaaS use without a separate deal. | vs MiniMax Speech: about half the languages, no instant voice cloning. |  | See pricing |  |
| 8 | [CosyVoice](https://www.versusref.com/tts/tools/cosyvoice/) (OSS) | Full-stack open TTS from Alibaba's speech team: 9 languages plus 18+ Chinese dialects, 150 ms streaming latency claim, instruction control of emotion/dialect/speed, and training + deployment scripts under Apache-2.0. | vs MiniMax Speech: about half the languages. |  | See pricing |  |
| 9 | [Deepgram Aura-2](https://www.versusref.com/tts/tools/deepgram-aura/) | Enterprise real-time voice-agent TTS | vs MiniMax Speech: about half the languages, ~75% cheaper, no instant voice cloning. |  | See pricing |  |
| 10 | [Dia / Dia2](https://www.versusref.com/tts/tools/dia/) (OSS) | Best-known open model for scripted two-speaker dialogue rather than narration - Apache-2.0, English-only, GPU-oriented. | vs MiniMax Speech: about half the languages. |  | See pricing |  |
| 11 | [ElevenLabs](https://www.versusref.com/tts/tools/elevenlabs/) | Premium AI voice platform for creators and developers | vs MiniMax Speech: ~70% faster, ~20% fewer languages. |  | From $6/mo |  |
| 12 | [F5-TTS](https://www.versusref.com/tts/tools/f5-tts/) (OSS) | The go-to research-grade voice-cloning model - actively maintained and broadly ported, with the classic code-vs-weights license split: commercial products must retrain or license around the CC-BY-NC checkpoints. |  |  | See pricing |  |
| 13 | [Fish Speech](https://www.versusref.com/tts/tools/fish-speech/) (OSS) | Top-tier expressive multilingual open-weights TTS whose license moved from permissive to research/non-commercial; commercial use requires a license from Fish Audio or their hosted API. | vs MiniMax Speech: ~2× the languages. |  | See pricing |  |
| 14 | [Google Cloud TTS](https://www.versusref.com/tts/tools/google-tts/) | Hyperscaler TTS with the broadest voice/language catalog | vs MiniMax Speech: ~95% cheaper, ~2× the languages. |  | See pricing |  |
| 15 | [GPT-SoVITS](https://www.versusref.com/tts/tools/gpt-sovits/) (OSS) | The de facto community standard for DIY voice cloning (60k GitHub stars), with a full WebUI covering dataset prep, ASR, training, and inference across five languages; MIT-licensed. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 16 | [Gradium](https://www.versusref.com/tts/tools/gradium/) | Real-time voice-agent infrastructure play: WebSocket-first streaming TTS in 5 European languages with instant cloning, on-device models (Phonon), and credit-based pricing. Active and well-funded (site announced funding extension to $100M, July 2026). | vs MiniMax Speech: about half the languages. |  | From $13/mo |  |
| 17 | [Grok TTS](https://www.versusref.com/tts/tools/grok-tts/) | Usage-priced hosted TTS from xAI, part of the Grok Voice stack (TTS, STT, and a speech-to-speech Voice Agent API), aimed at developers who want low-latency voice output alongside Grok models. | vs MiniMax Speech: ~85% cheaper, about half the languages. |  | See pricing |  |
| 18 | [Higgs Audio](https://www.versusref.com/tts/tools/higgs-audio/) (OSS) | Strong emotional/expressive open TTS from a well-funded lab - but 'Apache-2.0' only covers the repo code; V2 weights carry a 100k-MAU community license and the newer V3 is research/non-commercial. |  |  | See pricing |  |
| 19 | [Hume Octave TTS](https://www.versusref.com/tts/tools/hume-octave/) | Expressiveness-first TTS (Octave understands the meaning of the text it speaks); subscription plans gate commercial use, with Octave 2 (preview) adding ~100ms latency and 11 languages for realtime use. | vs MiniMax Speech: about half the languages, about half the latency. |  | From $3/mo |  |
| 20 | [IndexTTS-2](https://www.versusref.com/tts/tools/index-tts/) (OSS) | Emotion-controllable zero-shot voice cloning for production use, but under a custom Bilibili license (not OSI-approved) with scale caps and AI-training restrictions. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 21 | [Inworld TTS](https://www.versusref.com/tts/tools/inworld-tts/) | Cost-leader realtime TTS for voice agents and games; hybrid pay-as-you-go plus monthly credit plans that lower the per-1M-character rate as commitment grows; SOC 2 Type II with zero-data-retention posture. | vs MiniMax Speech: ~5× the languages, ~75% cheaper. |  | From $25/mo |  |
| 22 | [KittenTTS](https://www.versusref.com/tts/tools/kitten-tts/) (OSS) | The lightweight/edge option: ONNX inference, sub-100 MB downloads, 8 built-in voices - trades voice cloning and multilinguality for footprint. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 23 | [Kokoro](https://www.versusref.com/tts/tools/kokoro/) (OSS) | The lightweight quality-per-parameter champion of open TTS: Apache-2.0, easy to run anywhere, no voice cloning by design. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 24 | [Kyutai TTS](https://www.versusref.com/tts/tools/kyutai-tts/) (OSS) | Research-lab open TTS optimized for real-time streaming (Delayed Streams Modeling); Pocket TTS targets on-device/CPU deployment while the larger DSM TTS targets production streaming servers (Rust backend). | vs MiniMax Speech: about half the languages. |  | See pricing |  |
| 25 | [LMNT](https://www.versusref.com/tts/tools/lmnt/) | Speed-and-affordability challenger: a single fast model (Blizzard 2) with streaming-first APIs, unlimited voice clones on every tier, and flat subscription pricing with per-1K overage. | vs MiniMax Speech: about half the latency, ~25% fewer languages. |  | From $10/mo |  |
| 26 | [Maya1](https://www.versusref.com/tts/tools/maya1/) (OSS) | Apache 2.0 expressive English TTS you can run on a single 16GB+ GPU; stands out for natural-language voice design and 20+ inline emotion tags rather than audio-sample cloning, with real-time streaming via vLLM. | vs MiniMax Speech: about half the languages, no instant voice cloning. |  | See pricing |  |
| 27 | [MegaTTS3](https://www.versusref.com/tts/tools/megatts3/) (OSS) | Research-grade Apache-2.0 TTS whose practical cloning is gated: the WaveVAE encoder is not released, so users must submit audio to ByteDance channels to obtain pre-extracted speaker latents (.npy) for cloning. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 28 | [Murf API](https://www.versusref.com/tts/tools/murf/) | API arm of the Murf studio platform: 150+ voices in 35 languages, SSML support, word timestamps, and Falcon 2 aimed at high-concurrency voice agents at $0.01/1K characters. Note: Murf Studio subscription plans (murf.ai/pricing) are a separate product from API pay-as-you-go pricing and API characters are purchased separately. | vs MiniMax Speech: ~85% cheaper, ~70% cheaper, no instant voice cloning. |  | See pricing |  |
| 29 | [Narakeet](https://www.versusref.com/tts/tools/narakeet/) | Indie batch-content workhorse: aggregates a very broad voice/language catalog for voiceovers, audiobooks and video narration, priced per output minute (prepaid packs, no subscription) rather than per character; not aimed at real-time agent use. | vs MiniMax Speech: ~2.5× the languages, no instant voice cloning. |  | See pricing |  |
| 30 | [Neuphonic](https://www.versusref.com/tts/tools/neuphonic/) | Hybrid hosted + open-weights play: SSE/WebSocket streaming TTS API at app.neuphonic.com, and tiny CPU-only on-device models (NeuTTS-Air ~360M Apache-2.0, NeuTTS-Nano ~120M) in GGUF for phones/Raspberry Pi - privacy/edge-deployment angle. NOTE: the site's pricing page returned 404 at verification time; hosted-plan pricing treated as not published. | vs MiniMax Speech: about half the languages. |  | See pricing |  |
| 31 | [NVIDIA Magpie TTS](https://www.versusref.com/tts/tools/nvidia-magpie/) (OSS) | A small, GPU-efficient 9-language TTS checkpoint for teams already in the NVIDIA NeMo/Riva ecosystem; commercially usable open weights, but zero-shot voice cloning was removed from the open release and it caps generations at about 20 seconds. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 32 | [OpenAI TTS](https://www.versusref.com/tts/tools/openai-tts/) | Simple usage-based TTS inside a general AI platform | vs MiniMax Speech: ~75% cheaper, ~70% cheaper, no instant voice cloning. |  | See pricing |  |
| 33 | [Orpheus TTS](https://www.versusref.com/tts/tools/orpheus/) (OSS) | The 'LLM-as-TTS' approach under Apache-2.0 - expressive and clonable, but repo activity has slowed since mid-2025. | vs MiniMax Speech: about half the languages. |  | See pricing |  |
| 34 | [OuteTTS](https://www.versusref.com/tts/tools/outetts/) (OSS) | The llama.cpp-native option: runs via GGUF on CUDA/ROCm/Vulkan/Metal and even in the browser (Transformers.js). License splits by model: the 0.6B (Qwen3-based) is Apache-2.0, the 1B Llama-based flagship is CC-BY-NC-SA-4.0 (non-commercial). | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 35 | [Piper](https://www.versusref.com/tts/tools/piper/) (OSS) | The pragmatic embedded/self-hosted choice: no cloning or frills, just quick offline speech in many languages - note the license change from MIT (archived rhasspy/piper) to GPL-3.0 in the successor repo. | vs MiniMax Speech: no streaming audio output. |  | See pricing |  |
| 36 | [Qwen3-TTS](https://www.versusref.com/tts/tools/qwen3-tts/) (OSS) | Genuinely open weights (confirmed - not API-only since the Jan 2026 release): Apache-2.0 checkpoints on Hugging Face with an Alibaba Cloud DashScope API for hosted use. | vs MiniMax Speech: about half the languages. |  | See pricing |  |
| 37 | [Resemble AI](https://www.versusref.com/tts/tools/resemble/) | Security-first enterprise play: generation plus detection/verification in one platform, pay-as-you-go Flex credits, on-prem option, and the MIT-licensed open-source Chatterbox model family. |  |  | See pricing |  |
| 38 | [Rime](https://www.versusref.com/tts/tools/rime/) | Enterprise conversational TTS (IVR, contact centers, voice agents) emphasizing ultra-low latency models (Coda, Mist, Arcana) and self-hosted deployment; usage-based pricing with a single published rate. | vs MiniMax Speech: about half the latency, about half the price, no instant voice cloning. |  | See pricing |  |
| 39 | [Sarvam AI Bulbul](https://www.versusref.com/tts/tools/sarvam-bulbul/) | The go-to hosted TTS for Indian-language products (Hindi, Tamil, Telugu, Bengali and more) with REST, HTTP streaming and WebSocket APIs and prepaid INR pricing; not aimed at global multilingual coverage. | vs MiniMax Speech: about half the languages, no instant voice cloning. |  | See pricing |  |
| 40 | [Smallest.ai Waves](https://www.versusref.com/tts/tools/smallest-ai/) | Speed- and price-led challenger from an India-focused voice AI startup; Waves is the speech-model API layer (Lightning TTS, Pulse STT), sold pay-as-you-go with $10 free credits, with HIPAA/SOC2/on-prem reserved for the Enterprise plan. | vs MiniMax Speech: about half the languages, about half the latency. |  | See pricing |  |
| 41 | [Speechify API](https://www.versusref.com/tts/tools/speechify-api/) | Developer platform spun out of the Speechify brand: transparent tiered pricing ($10-$499/mo plus per-1M overage), streaming-native Simba 3.2, and a bundled voice-agents product with flat per-minute rates. | vs MiniMax Speech: ~90% cheaper, ~25% fewer languages. |  | From $10/mo |  |
| 42 | [Speechmatics TTS](https://www.versusref.com/tts/tools/speechmatics-tts/) | Budget low-latency English TTS for voice agents; priced at a fraction of premium voice APIs, but currently English-only with a small voice set and no SSML or voice cloning. | vs MiniMax Speech: about half the languages, ~90% cheaper, no instant voice cloning. |  | See pricing |  |
| 43 | [Step-Audio](https://www.versusref.com/tts/tools/step-audio/) (OSS) | Differentiates on iterative audio *editing* (emotion, style, breathing, laughter, sighs, polyphone pinyin control) rather than plain synthesis; Apache-2.0 with training code (SFT/DPO/GRPO) and vLLM support. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |
| 44 | [Supertone API](https://www.versusref.com/tts/tools/supertone/) | Expressive character voices (games, content, Korean/Japanese/English markets first, now 31 languages via Supertonic 3) on cheap credit-based subscriptions from $2.99/month; uniquely pairs the hosted API with the open OpenRAIL-M Supertonic model for on-device synthesis. | vs MiniMax Speech: ~25% fewer languages. |  | From $2.99/mo |  |
| 45 | [Typecast](https://www.versusref.com/tts/tools/typecast/) | Expressiveness play from a Korean AI-actor studio: strongest on emotion prompts/presets and character voices for content and conversational AI; separate consumer studio subscription ($8.99+) and developer API plans (Free/Lite/Plus). | vs MiniMax Speech: ~30% cheaper, ~20% faster. |  | From $15/mo |  |
| 46 | [Unreal Speech](https://www.versusref.com/tts/tools/unreal-speech/) | Pure price play: markets itself as up to 11x cheaper than ElevenLabs with a simple 3-endpoint API (stream/speech/synthesisTasks); smaller voice/language catalog and no voice cloning documented. | vs MiniMax Speech: about half the languages, ~20% slower, no instant voice cloning. |  | From $49/mo |  |
| 47 | [VibeVoice](https://www.versusref.com/tts/tools/vibevoice/) (OSS) | The open long-form/multi-speaker specialist - MIT weights, but Microsoft pulled the TTS code from the repo in Sept 2025 and frames the models as research-only. | vs MiniMax Speech: about half the languages, no instant voice cloning. |  | See pricing |  |
| 48 | [Voxtral (open weights)](https://www.versusref.com/tts/tools/voxtral-open/) (OSS) | A frontier-lab open-weight TTS you can run on a single 16GB GPU; the CC BY-NC license makes it evaluation/research-only, with Mistral's paid API as the commercial route. | vs MiniMax Speech: about half the languages. |  | See pricing |  |
| 49 | [Voxtral TTS](https://www.versusref.com/tts/tools/voxtral-tts/) | Open-weights-friendly voice cloning TTS from a frontier AI lab | vs MiniMax Speech: ~85% cheaper, about half the languages. |  | See pricing |  |
| 50 | [WellSaid](https://www.versusref.com/tts/tools/wellsaid/) | Enterprise voiceover specialist: polished Studio product for L&D/marketing narration with an API on the side; API pricing is contact-sales, and compliance (SOC 2 Type II, GDPR) and ethical voice sourcing are the pitch. | vs MiniMax Speech: no instant voice cloning. |  | From $10/mo |  |
| 51 | [XTTS v2 (Coqui)](https://www.versusref.com/tts/tools/xtts/) (OSS) | The legacy standard for open voice cloning: dormant upstream since Coqui's Jan 2024 shutdown (community fork idiap/coqui-ai-TTS carries maintenance), and CPML weights bar commercial use. | vs MiniMax Speech: about half the languages, no streaming audio output. |  | See pricing |  |

## How the top MiniMax Speech alternatives compare

Beyond the ranked cards: how the top 6 MiniMax Speech alternatives place in the Text-to-speech APIs field, where each one wins in our published verdicts, and what MiniMax Speech still holds over it.

### 1. Fish Audio

Among the 46 text-to-speech tools we track, Fish Audio has the 5th-widest language coverage and the 3rd-cheapest flagship rate - a fit for multilingual and localization projects and cost-sensitive, high-volume work.
In our published verdicts, Fish Audio beats MiniMax Speech for audiobooks, dubbing, self-hosted and voice agents.
The reverse angle matters too - MiniMax Speech vs Fish Audio: ~565% pricier, ~150% slower.
Published pricing starts at $15 per 1M characters, verified July 2026.

**Why teams switch (Audiobooks):** For audiobook production, per-character cost is decisive: Fish Audio charges $15 per 1M UTF-8 bytes versus MiniMax Speech at $100 per 1M characters, making Fish Audio roughly 6x cheaper at scale. MiniMax also caps input at under 10,000 characters per request, which is a significant friction point for book-length content. Fish Audio additionally supports 83 languages versus 40 for MiniMax, and offers a self-host option for studios wanting on-premises control. MiniMax has pronunciation dictionaries, but cost and input-length constraints outweigh that advantage for long-form narration. [Audiobooks verdict](https://www.versusref.com/tts/fish-audio-vs-minimax-speech/)

**Overall verdict:** Fish Audio wins 4 of 6 use case verdicts versus MiniMax Speech's 2. It also leads on core metrics: flagship pricing at 15 $/1M UTF-8 bytes versus 100 $/1M chars, TTFB of 100 ms versus 250 ms, 83 languages versus 40, and a self-host option MiniMax lacks entirely. MiniMax retains advantages in concurrency (60 RPM base vs 5) and developer tooling extras like word timestamps. Even so, Fish Audio is the safer default for most buyers across price, latency, and deployment flexibility.

Fish Audio cost: Usage-priced at $15 per 1M characters (≈ $0.0143 per audio-minute), verified Jul 20, 2026.

| Monthly volume | $ / 1M chars | $ / audio-min | Monthly bill |
| --- | --- | --- | --- |
| 200K chars/mo | $15 | $0.0143 | $3 |
| 2M chars/mo | $15 | $0.0143 | $30 |
| 20M chars/mo | $15 | $0.0142 | $300 |
[Full Fish Audio vs MiniMax Speech comparison](https://www.versusref.com/tts/fish-audio-vs-minimax-speech/) · [Fish Audio review](https://www.versusref.com/tts/tools/fish-audio/)

### 2. Amazon Polly

Among the 10 text-to-speech tools we track, Amazon Polly has the 1st-cheapest fast-model rate and the 10th-widest language coverage - a fit for multilingual and localization projects.
The reverse angle matters too - MiniMax Speech vs Amazon Polly: ~1400% pricier, ~235% pricier, adds instant voice cloning.
Published pricing starts at $30 per 1M characters, verified July 2026.

Amazon Polly cost: Usage-priced at $30 per 1M characters (≈ $0.0285 per audio-minute), verified Jul 20, 2026.

| Monthly volume | $ / 1M chars | $ / audio-min | Monthly bill |
| --- | --- | --- | --- |
| 200K chars/mo | $30 | $0.0285 | $6 |
| 2M chars/mo | $30 | $0.0285 | $60 |
| 20M chars/mo | $30 | $0.0285 | $600 |
[Amazon Polly review](https://www.versusref.com/tts/tools/amazon-polly/)

### 3. Azure Speech

Among the 46 text-to-speech tools we track, Azure Speech has the 3rd-widest language coverage and the 6th-cheapest flagship rate - a fit for multilingual and localization projects and cost-sensitive, high-volume work.
Before switching, weigh what stays behind - MiniMax Speech vs Azure Speech: ~355% pricier, ~300% pricier.
Its published rate is $22 per 1M characters, verified July 2026.

Azure Speech cost: Usage-priced at $22 per 1M characters (≈ $0.0209 per audio-minute), verified Jul 20, 2026.

| Monthly volume | $ / 1M chars | $ / audio-min | Monthly bill |
| --- | --- | --- | --- |
| 200K chars/mo | $22 | $0.0209 | $4.40 |
| 2M chars/mo | $22 | $0.0209 | $44 |
| 20M chars/mo | $22 | $0.0209 | $440 |
[Azure Speech review](https://www.versusref.com/tts/tools/azure-speech/)

### 4. CAMB.AI

Among the 46 text-to-speech tools we track, CAMB.AI has the 2nd-widest language coverage - a fit for multilingual and localization projects.
Before switching, weigh what stays behind - MiniMax Speech vs CAMB.AI: about half the languages.
[CAMB.AI review](https://www.versusref.com/tts/tools/camb-ai/)

### 5. Cartesia

Among the 46 text-to-speech tools we track, Cartesia has the 9th-widest language coverage and the 3rd-fastest time-to-first-byte - a fit for multilingual and localization projects and real-time, conversational apps.
Before switching, weigh what stays behind - MiniMax Speech vs Cartesia: ~180% slower.
[Cartesia review](https://www.versusref.com/tts/tools/cartesia/)

### 6. Chatterbox

Among the 46 text-to-speech tools we track, Chatterbox has the 18th-widest language coverage.
Before switching, weigh what stays behind - MiniMax Speech vs Chatterbox: ~2× the languages, adds streaming audio output.
[Chatterbox review](https://www.versusref.com/tts/tools/chatterbox/)

Source: https://www.versusref.com/tts/alternatives/minimax-speech/